ECS Overview and Architecture

April 2024

H14071.25

White Paper

Abstract
This document provides a technical overview and design of the Dell
ECS software-defined cloud-scale object storage platform.

Dell Technologies
 Copyright

The information in this publication is provided as is. Dell Inc. makes no representations or warranties of any kind with respect
to the information in this publication, and specifically disclaims implied warranties of merchantability or fitness for a particular
purpose.
Use, copying, and distribution of any software described in this publication requires an applicable software license.
Copyright © 2015-2023 Dell Inc. or its subsidiaries. All Rights Reserved. Published in the USA April 2024 H14071.25.
Dell Inc. believes the information in this document is accurate as of its publication date. The information is subject to change
without notice.

2 ECS Overview and Architecture

 Contents

Contents
Executive summary.......................................................................................................................4

Value of ECS ..................................................................................................................................6

Architecture ...................................................................................................................................8

Appliance hardware models .......................................................................................................30

Network separation .....................................................................................................................33

Security ........................................................................................................................................34

Data integrity and protection ......................................................................................................42

Deployment..................................................................................................................................44

Storage protection overhead ......................................................................................................56

Conclusion...................................................................................................................................58

Resources ....................................................................................................................................59

ECS Overview and Architecture 3

 Executive summary

Executive summary

Introduction Organizations require options for consuming public cloud services with the reliability and
control of a private-cloud infrastructure. Dell ECS is a software-defined, cloud-scale,
object storage platform that delivers S3, Atmos, CAS, Swift and NFSv3, storage services
on a single, modern platform.

With ECS, administrators can easily manage globally distributed storage infrastructure
under a single global namespace that provides strong consistency across sites. ECS core
components are layered for flexibility and resiliency. Each layer is abstracted and
independently scalable with high availability.

Simple RESTful API access for storage services is being embraced by developers. Use of
HTTP semantics like GET and PUT simplifies the application logic required when
compared with traditional, but familiar, path-based file operations. In addition, ECS’s
underlying storage system is strongly consistent, which means it can guarantee an
authoritative response. Applications that are required to guarantee authoritative delivery of
data can do so without complex code logic by using ECS.

Audience This paper is intended for anyone interested in understanding the value and architecture
of ECS. It aims to provide context with links to additional information.

Scope This document provides an overview of the Dell ECS object storage platform. It details the
ECS design architecture and core components such as the storage services and data
protection mechanisms.

This document focuses primarily on ECS architecture. It does not cover installation,
administration, and upgrade procedures for ECS software or hardware. It also does not
cover specifics on using and creating applications with the ECS APIs.

Updates to this document are done periodically and generally coincide with major
releases or new features.

Revisions Part
Date number/ Description
revision

December 2015 Initial release

May 2016 Updated for 2.2.1

September 2016 Updated for 3.0

August 2017 Updated for 3.1

March 2018 Updated for 3.2

September 2018 Updated for Gen3 Hardware

February 2019 Updated for 3.3

September 2019 Updated for 3.4

4 ECS Overview and Architecture

 Executive summary

Part
Date number/ Description
revision

February 2020 Updated ECSDOC-628 changes

May 2020 Updated for 3.5

November 2020 Updated for 3.6

February 2021 Updated for 3.6.1

August 2021 Updated for 3.6.2 and compression mechanism

December 2021 Updated template

February 2022 Updated for 3.7

December 2022 H14071.22 Updated for 3.8.0.1

March 2023 H14071.23 Minor updates

November 2023 H14071.24 Updated 20TB with EX5000 and remove the intermix
between HDD and AFA cluster

April 2024 H14071.25 Updated for 3.8.1

We value your Dell Technologies and the authors of this document welcome your feedback on this
feedback document. Contact the Dell Technologies team by email.

Author: Jarvis Zhu

Note: For links to other documentation for this topic, see the ECS Info Hub.

ECS Overview and Architecture 5

 Value of ECS

Value of ECS
ECS provides significant value for enterprises and service providers seeking a platform
architected to support rapid data growth. The main advantages and features of ECS that
enable enterprises to globally manage and store distributed content at scale include:

Cloud Scale - ECS is an object storage platform for both traditional and next-gen
workloads. ECS’s software-defined layered architecture promotes limitless scalability.
Feature highlights are:
• Globally distributed object infrastructure
• Exabyte+ scale without limits on storage pool, cluster, or federated environment
capacity
• No limits exist on the number of objects in a system, namespace, or bucket
• Efficient at both small and large file workloads with no limits to object size
Flexible Deployment - ECS has unmatched flexibility with features such as:
• Appliance deployment
• Software-only deployment with support for certified or custom industry standard
hardware
• Multiprotocol support: Object (S3, Swift, Atmos, CAS) and File ( NFSv3)
• Multiple workloads: Modern apps and traditional apps
• Secondary storage for Data Domain Cloud Tier and Isilon using CloudPools
• Non-disruptive upgrade paths to current generation ECS models
Enterprise Grade - ECS provides customers more control of their data assets with
enterprise class storage in a secure and compliant system with features such as:
• Data-at-rest (D@RE) with key rotation and external key management.
• Encrypted inter-site communication
• Reporting, policy and event based record retention and platform hardening for
SEC Rule 17a-4(f) compliance including advanced retention management such
as litigation hold and min-max governance
• Compliance with SuSE Linux Enterprise Server (SLES) Security Technical
Implementation Guide (STIG) hardening guidelines
• Authentication, authorization, and access controls with Active directory and
LDAP
• Integration with monitoring and alerting infrastructure (SNMP traps and
SYSLOG)
• Enhanced enterprise capabilities (multi-tenancy, capacity monitoring and
alerting)

6 ECS Overview and Architecture

 Value of ECS

TCO Reduction - ECS can dramatically reduce Total Cost of Ownership (TCO) relative to
both traditional storage and public cloud storage. It even offers a lower TCO than tape for
long-term retention. Features include:
• Global namespace
• Small and large file performance
• Seamless Centera migration
• Fully compliant with Atmos REST
• Low management overhead
• Small data center footprint
• High storage utilization
The design of ECS is optimized for the following primary use cases:
• Modern Applications - ECS designed for modern development such as for
next-gen web, mobile, and cloud applications. Application development is
simplified with strongly-consistent storage. Along with multi-site, simultaneous
multi-user read/write access, as the ECS capacity changes and grows,
developers never need to recode their apps.
• Secondary Storage - ECS is used as secondary storage to free up primary
storage of infrequently accessed data, while also keeping it reasonably
accessible. Examples are policy-based tiering products such as Data Domain
Cloud Tier and Isilon CloudPools. GeoDrive, a Windows-based application,
gives Windows systems direct access to ECS to store data.
• Geo-Protected Archive - ECS serves as a secure and affordable on-premises
cloud for archival and long-term retention purposes. Using ECS as an archive
tier can significantly reduce primary storage capacities. To allow for better
storage efficiencies for cold archive use cases a 10+2 erasure coding (EC)
scheme is available in addition to the default of 12+4.
• Global Content Repository - Unstructured content repositories containing data
such as images and videos are often stored in high cost storage systems
making it impossible for businesses to cost-effectively manage massive data
growth. ECS enables consolidation of multiple storage systems into a single,
globally accessible and efficient content repository.
• Storage for Internet of Things - The Internet of Things (IoT) offers a new
revenue opportunity for businesses who can extract value from customer data.
ECS offers an efficient IoT architecture for unstructured data collection at
massive scale. With no limits on the number of objects, the size of objects or
custom metadata, ECS is the ideal platform to store IoT data. ECS can also
streamline some analytic workflows by allowing data to be analyzed directly on
the ECS platform without requiring time consuming extract, transform and load
(ETL) processes. Hadoop clusters can run queries using data stored on ECS by
another protocol API such as S3 or NFS.
• Video Surveillance Evidence Repository - In contrast to IoT data, video
surveillance data has a much smaller object storage count, but a much higher
capacity footprint per file. While data authenticity is important, data retention is
not as critical. ECS can be a low-cost landing area or secondary storage

ECS Overview and Architecture 7

 Architecture

location for this data. Video management software can leverage the rich custom
metadata capabilities for tagging files with important details like camera
location, retention requirement and data protection requirement. Also, metadata
can be used to set the file to a read-only status to ensure a chain of custody on
the file.
• Data lakes and Analytics - Data and analytics have become a competitive
differentiator and a primary source of value generation for organizations.
However, transforming data into a valuable corporate asset is a complex topic
that can easily entail the use of dozens of technologies, tools, and
environments. ECS provide a set of services to help customer collecting,
storing, governing, and analyzing data at any scale.

Architecture

Introduction ECS is architected with a few core design principles, such as global namespace with
strong consistency; scale-out capability, secure multi-tenancy; and superior performance
for both small and large objects. ECS is built as a completely distributed system following
the principle of cloud applications, where every function in the system is built as an
independent layer. With this design, each layer is horizontally scalable across all nodes in
the system. Resources are distributed across all nodes to increase availability and share
the load.

This section will go in-depth into the ECS architecture and design of the software and
hardware.

Architecture ECS is deployed on a set of qualified industry standard hardware or as a turnkey storage
overview appliance. The main components of ECS are the:
• ECS Portal and Provisioning Services - API-based WebUI and CLI for self-
service, automation, reporting, and management of ECS nodes. This layer also
handles licensing, authentication, multi-tenancy, and provisioning services such
as namespace creation.
• Data Services - Services, tools and APIs to support object and file access to
the system.
• Storage Engine - Core service responsible for storing and retrieving data,
managing transactions, and protecting and replicating data locally and between
sites.
• Fabric - Clustering service for health, configuration, and upgrade management
and alerting.
• Infrastructure - SUSE Linux Enterprise Server 12 for the base operating
system in the turnkey appliance or qualified Linux operating systems for industry
standard hardware configuration.
• Hardware - A turnkey appliance or qualified industry standard hardware.
The following figure shows a graphical view of these layers which are described in detail
in the sections that follow.

8 ECS Overview and Architecture

 Architecture

Figure 1. ECS architecture layers

ECS portal and Storage administrators manage ECS using the ECS Portal and provisioning services.
provisioning ECS provides a web-based GUI (WebUI) to manage, license, and provision ECS nodes.
services The portal has comprehensive reporting capabilities that include:
• Capacity utilization per site, storage pool, node, and disk.
• Performance monitoring on latency, throughput, and replication progress.
• Diagnostic information, such as node and disk recovery status.
The ECS dashboard provides overall system-level health and performance information.
This unified view enhances overall system visibility. Alerts notify users about critical
events, such as capacity limits, quota limits, disk or node failures or software failures.
ECS also provides a command-line interface to install, upgrade, and monitor ECS. Access
to nodes for command-line usage is done using SSH. The following figure shows the ECS
dashboard:

ECS Overview and Architecture 9

 Architecture

Figure 2. ECS Web UI dashboard

Detailed performance reporting is available in the UI under the Advance Monitoring folder.
The reports are displayed in a Grafana dashboard. There are filters available to drill into
specified Namespaces, Protocols, or Nodes. The following figure shows an example of an
S3 protocol performance report:

10 ECS Overview and Architecture

 Architecture

Figure 3. Advanced monitoring visualization using Grafana

ECS can also be managed using RESTful APIs. The management API allows users to
administer ECS within their own tools, scripts, and new or existing applications. The ECS
web UI and command-line tools are built using the ECS REST Management APIs.

ECS supports the following event notification servers which can be set using the web UI,
API, or CLI:
• SNMP (Simple Network Management Protocol) servers
• Syslog servers
The ECS Administrator’s Guide has more information and details on configuring
notification services.

Data services Standard object and file methods are used to access ECS storage services. For S3,
Atmos and Swift, RESTful APIs over HTTP are used for access. For Content Addressable
Storage (CAS), a proprietary access method/SDK is used. ECS natively supports all the
NFSv3 procedures except for LINK. ECS buckets can now be accessed by S3a.

ECS provides multi-protocol access where data ingested through one protocol can be
accessed through others. This means that data can be ingested through S3 and modified

ECS Overview and Architecture 11

 Architecture

through NFSv3 or Swift, or vice versa. There are some exceptions to multi-protocol
access due to protocol semantics and representations of protocol design. The following
table highlights the access methods and which protocols interoperate.

Table 1. ECS supported data services and protocol interoperability

Protocol Supported Interoperability

Object S3 Additional capabilities like Byte NFS, Swift, S3a

Range Updates and Rich ACLS

Atmos Version 2.0 NFS (path-based objects only and

not object ID style based)

Swift V2 APIs and Swift and Keystone NFS, S3

v3 Authentication

CAS SDK v3.1.544 or later N/A

File NFS NFSv3 S3, Swift, Atmos (path-based objects

only and not object ID style based)

Data services, which are also referred to as head services, are responsible for taking
client requests, extracting required information, and passing it to the storage engine for
further processing. All head services are combined to a single process, dataheadsvc,
running inside the infrastructure layer. This process is further encapsulated within a
Docker container named object-main that runs on every node within ECS. Infrastructure
covers Docker in more detail. ECS protocol service port requirements, such as port 9020
for S3 communication, are available in the latest ECS Security Configuration Guide.

Object
ECS supports S3, Atmos, Swift, and CAS APIs for object access. Except for CAS, objects
or data are written, retrieved, updated, and deleted using HTTP or HTTPS calls of GET,
POST, PUT, DELETE, and HEAD. For CAS, standard TCP communication and specific
access methods and calls are used.

ECS provides a facility for metadata search for objects using a rich query language. This
is a powerful feature of ECS that allows S3 object clients to search for objects within
buckets using system and custom metadata. While search is possible using any
metadata, by searching on metadata that has been specifically configured to be indexed
in a bucket, ECS can return queries quicker, especially for buckets with billions of objects.

Metadata search with tokenization allows the customer to use metadata search to search
for objects that have a specific metadata value within an array of metadata values. The
method must be chosen when the bucket is created. It can be included as an option when
creating the bucket through the S3 create bucket API, and include the header x-emc-
metadata-search-tokens: true in the request.

Up to thirty user-defined metadata fields can be indexed per bucket. Metadata is specified
at the time of bucket creation. Metadata search feature can be enabled on buckets with
server-side encryption enabled; however, any indexed user metadata attribute used as a
search key will not be encrypted.

Note: There is a performance impact when writing data in buckets configured to index metadata.
The impact to operations increases as the number of indexed fields increases. Impact to

12 ECS Overview and Architecture

 Architecture

performance needs careful consideration on choosing if to index metadata in a bucket, and if so,
how many indexes to maintain.

For CAS objects, CAS query API provides similar ability to search for objects based on
metadata that is maintained for CAS objects which does not need to be enabled explicitly.

For more information about about ECS APIs and APIs for metadata search, see the latest
ECS Data Access Guide. For Atmos and S3 SDKs refer to the GitHub site Dell Data
Services SDK or Dell ECS. For CAS refer to the Centera Community site. Access to
numerous examples, resources, and assistance for developers can be found in the ECS
Community.

Client applications such as S3 Browser and Cyberduck provide a way to quickly test or
access data stored in ECS. ECS Test Drive is freely provided by Dell which allows access
to a public facing ECS system for testing and development purposes. After registering for
ECS Test Drive, REST endpoints are provided with user credentials for each of the object
protocols. Anyone can use ECS Test drive to test their S3 API application.

Note: Only the number of metadata that can be indexed per bucket is limited to thirty in ECS.
There is no limitation to the total number of custom metadata stored per object, only the number
indexed for fast lookup.

Hadoop S3A support

ECS supports the Hadoop S3A client for storing Hadoop data. S3A is an open source
connector for Hadoop, based on the official Amazon Web Services (AWS) SDK. It was
created to address storage scaling and cost problems that many Hadoop admin were
having with HDFS. Hadoop S3A connects Hadoop clusters to any S3 compatible object
store whether in the public, hybrid, or on-premises cloud.

Note: S3A support is available on Hadoop 3.1.1.

ECS Overview and Architecture 13

 Architecture

Figure 4. Hadoop and ECS architecture

As shown in the preceding figure, when the Hadoop cluster is set up on traditional HDFS,
its S3A configuration points to the ECS Object data to do all the HDFS activity. On each
Hadoop HDFS node, any traditional Hadoop component would use the Hadoop’s S3A
client to perform the HDFS activity.

Hadoop configuration analysis using ECS Service Console

The ECS Service Console (SC) can read and interpret your Hadoop configuration
parameters with respect to connections to ECS for S3A. Also, SC provides a function,
Get_Hadoop_Config that reads the Hadoop cluster configuration and checks S3A settings
for typos, errors, and values. Contact ECS support team for assistance with installing ECS
SC.

Privacera implementation with Hadoop S3A

Privacera is a third-party vendor that has implemented a Hadoop client-side agent and
integration with Ambari for S3 (AWS and ECS) granular security. Although Privacera
supports Cloudera Distribution of Hadoop (CDH), Cloudera (another third-party vendor)
does not support Privacera on CDH.

Note: CDH users must use ECS IAM security services. If you want secure access to S3A without
using ECS IAM, contact the support team.

See the latest ECS Data Access Guide for more information about S3A support.

14 ECS Overview and Architecture

 Architecture

Hadoop S3A security

ECS IAM allows the Hadoop administrator to setup access policies to control access to
S3A Hadoop data. Once the access policies are defined, there are two user access
options for Hadoop administrators to configure:
• IAM Users/Groups
▪ Create IAM groups that attach to policies
▪ Create IAM users that are members of an IAM group
• SAML Assertions (Federated Users)
▪ Create IAM roles that attach to policies
▪ Configure CrossTrustRelationship between Identity Provider (AD FS) and ECS
that map AD groups to IAM roles
ECS admin and Hadoop admin need to work together to pre-define appropriate policies.
The fictional examples that follow outline three types of Hadoop users that we will create
policies for. They are:
• Hadoop Administrator - do all operations, except create bucket and delete
bucket
• Hadoop Power User - do all operations except create bucket, delete bucket
and delete objects
• Hadoop Read Only User - only list and read objects
For more information about ECS IAM, see ECS IAM.

NFS
ECS includes native file support with NFSv3. The main features for the NFSv3 file data
service include:
• Global namespace - File access from any node at any site.
• Global locking - In NFSv3 locking is advisory only. ECS supports compliant
client implementations that allow for shared and exclusive, range-based and
mandatory locks.
• Multiprotocol access - Access to data using different protocol methods.
NFS exports, permissions and user group mappings are created using the WebUI or API.
NFSv3 compliant clients mount exports using namespace and bucket names. Here is a
sample command to mount a bucket:

mount –t nfs –o vers=3 s3.dell.com:/namespace/bucket

To achieve client transparency during a node failure, a load balancer is recommended for
this workflow.

ECS has tightly integrated the other NFS server implementations, such as lockmgr, statd,
nfsd, and mountd, hence, these services are not dependent on the infrastructure layer
(host operating system) to manage. NFSv3 support has the following features:
• No design limits on the number of files or directories.

ECS Overview and Architecture 15

 Architecture

• File write size can be up to 16TB.

• Ability to scale across up to 8 sites with a single global namespace/export.
• Support for Kerberos and AUTH_SYS authentication.
NFS file services process NFS requests coming from clients; however, data is stored as
objects within ECS. An NFS file handle is mapped to an object id. Since the file is
basically mapped to an object, NFS has features like the object data service, including:
• Quota management at the bucket level.
• Encryption at the object level.
• Write-Once-Read-Many (WORM) to the bucket level.
▪ WORM is implemented using Auto Commit period during new bucket creation.
▪ WORM is only applicable to non-compliant buckets.
Connectors and gateways
Several third-party software products can access ECS object storage. Independent
software vendors (ISVs) such as Panzura, Ctera, and Syncplicity create a layer of
services that offer client access to ECS object storage using traditional protocols such as
SMB/CIFS, NFS, and iSCSI. Organizations can also access or upload data to ECS
storage with the following Dell products:
• PowerScale CloudPools - Policy-based tiering of data to ECS from
PowerScale.
• Data Domain Cloud Tier - Automated native tiering of deduplicated data to
ECS from Data Domain for long-term retention. Data Domain Cloud Tier
provides a secure and cost-effective solution to encrypt data in the cloud with a
reduced storage footprint and network bandwidth.
• GeoDrive - ECS stub-based storage service for Microsoft® Windows®
desktops and servers.

Storage engine At the core of ECS is the storage engine. The storage engine layer contains the main
components responsible for processing requests and storing, retrieving, protecting, and
replicating data.

This section describes the design principles and how data is represented and handled
internally.

Storage services
The ECS storage engine includes the services shown in the following figure:

16 ECS Overview and Architecture

 Architecture

Figure 5. Storage engine services

The services of the Storage Engine are encapsulated within a Docker container that runs
on every ECS node to provide a distributed and shared service.

Data
The primary types of data stored in ECS can be summarized as follows:
• Data - Application- or user-level content stored such as an image. Data is used
synonymously with object, file, or content. Applications may store an unlimited
amount of custom metadata with each object. The storage engine writes data
and associated application-provided custom metadata together in a logical
repository. Custom metadata is a robust feature of modern storage systems that
provide further information or categorization of the data being stored. Custom
metadata is formatted as key-value pairs and provided with write requests.
• System metadata - System information and attributes relating to user data and
system resources. System metadata can be broadly categorized as follows:
▪ Identifiers and descriptors - A set of attributes used internally to identify
objects and their versions. Identifiers are either numeric ids or hash values
which are not of use outside the ECS software context. Descriptors define
information such as type of encoding.
▪ Encryption keys in encrypted format - Data encryption keys are considered
system metadata. They are stored in encrypted form inside the core directory
table structure.
▪ Internal flags - A set of indicators used to track if byte range updates or
encryption are enabled, and to coordinate caching and deletion.
▪ Location information - Attribute set with index and data location such as byte
offsets.
▪ Timestamps - Attribute set that tracks time such as for object create or update.
▪ Configuration/tenancy information - Namespace and object access control.

ECS Overview and Architecture 17

 Architecture

Data and system metadata are written in chunks on ECS. An ECS chunk is a 128MB
logical container of contiguous space. Each chunk can have data from different objects,
as shown below in the following figure. ECS uses indexing to keep track of all the parts of
an object that may be spread across different chunks and nodes.

Figure 6. 128MB chunk storing data of three objects

Chunks are written in an append-only pattern. The append-only behavior means that an
application’s request to modify or update an existing object will not modify or delete the
previously written data within a chunk, but rather the new modifications or updates will be
written in a new chunk. Therefore, no locking is required for I/O and no cache invalidation
is required. The append-only design also simplifies data versioning. Old versions of the
data are maintained in previous chunks. If S3 versioning is enabled and an older version
of the data is needed, it can be retrieved or restored to a previous version using the S3
REST API.

Data integrity and protection explains how data is protected at the chunk level.

Data management
ECS has a built-in snappy compression mechanism. The granularity is 2MB for small
objects and 128MB for large objects. ECS employs a smart compression logic where it
only compresses data that is compressible, saving resources from trying to compress
already compressed or in-compressible data (such as encrypted data or video files). If
more sophisticated compression is required, the ECS Java SDK supports client side ZIP
and LZMA.

ECS uses a set of logical tables to store information relating to the objects. Key-value
pairs are eventually stored on disk in a B+ tree for fast indexing of data locations. By
storing the key-value pair in a balanced, searched tree like a B+ tree, the location of the
data and metadata can be accessed quickly. ECS implements a two-level log-structured
merge tree where there are two tree-like structures; a smaller tree is in memory (memory
table) and the main B+ tree resides on disk. Lookup of key-value pairs occurs in memory
first subsequently at the main B+ tree on disk if needed. Entries in these logical tables are
first recorded in journal logs and these logs are written to disks in triple-mirrored chunks.
The journals are used to track transactions not yet committed to the B+ tree. After each
transaction is logged into a journal, the in-memory table is updated. Once the table in the
memory becomes full or after a certain period, its merge sorted or dumped to B+ tree on
disk. The number of journal chunks used by the system is insignificant when compared to
B+ tree chunks. The following figure illustrates this process:

18 ECS Overview and Architecture

 Architecture

Figure 7. Memory table dumped to B+ tree

The following table shows the information stored in the Object (OB) table. The OB table
contains the names of objects and their chunk location at a certain offset and length within
that chunk. In this table, the object name is the key to the index and the value is the chunk
location. The index layer within the Storage Engine is responsible for the object name-to-
chunk mapping.

Table 2. Object table entries

Object name Chunk location

ImgA C1:offset:length

FileB • C2:offset:length
• C3:offset:length

The chunk table (CT) records the location for each chunk, as detailed in the following
table:

Table 3. Chunk table entries

Chunk ID Location

C1 • Node1:Disk1:File1:Offset1:Length
• Node2:Disk2:File1:Offset2:Length
• Node3:Disk2:File6:Offset:Length

ECS was designed to be a distributed system such that storage and access of data are
spread across all nodes. Tables used to manage object data and metadata grow over
time as the storage is used and grows. The tables are divided into partitions and assigned
to different nodes where each node becomes the owner of the partitions it is hosting for
each of the tables. To get the location of a chunk, for example, the Partition Records table
(PR) is queried for owner node which has knowledge of the chunk location. A basic PR
table is illustrated in the following table:

ECS Overview and Architecture 19

 Architecture

Table 4. Partition records table entries

Partition ID Owner

P1 Node 1

P2 Node 2

P3 Node 3

If a node goes down, other nodes take ownership of its partitions. The partitions are
recreated by reading the B+ tree root and replaying the journals stored on disk. The
following figure shows the failover of partition ownership:

Figure 8. Failover of partition ownership

Data flow
Storage services are available from any node. Data is protected by distributed EC
segments across drives, nodes, and racks. ECS runs a checksum function and stores the
result with each write. If the first few bytes of data are compressible ECS will compress
the data. With reads, data is decompressed, and its stored checksum is validated. Here is
an example of a data flow for a write in five steps:
1. Client sends object create request to a node.
2. Node servicing the request writes the new object’s data into a repo (short for
repository) chunk.
3. On successful write to disk a PR transaction occurs to enter name and chunk
location.
4. The partition owner records the transaction in journal logs.
5. Once the transaction has been recorded in the journal logs, an acknowledgment is
sent to the client.
The following figure shows an example of the data flow for a read for hard disk drive
architecture like Gen2 and EX300, EX500, EX3000, and EX5000.
1. A read object request is sent from client to Node 1.

20 ECS Overview and Architecture

 Architecture

2. Node 1 uses a hash function using the object name to determine which node is the
partition owner of the logical table where this object information resides. In this
example, Node 2 is owner and thus Node 2 will do a lookup in the logical tables to
get location of chunk. In some cases, the lookup can occur on two different nodes,
for instance when the location is not cached in logical tables of Node 2.
3. From the previous step, location of chunk is provided to Node 1 who will then issue
a byte offset read request to the node that holds the data, Node 3 in this example,
and will send data to Node 1.
4. Node 1 sends data to requesting client.

Figure 9. Read data flow for hard disk drive architecture

The following figure Figure 10shows an example the data flow for a read for all flash
architecture like EXF900.
1. A read object request is sent from client to Node 1.
2. Node 1 uses a hash function using the object name to determine which node is the
partition owner of the logical table where this object information resides. In this
example, Node 2 is owner and thus Node 2 will do a lookup in the logical tables to
get location of chunk. In some cases, the lookup can occur on two different nodes,
for instance when the location is not cached in logical tables of Node 2.
3. From the previous step, location of chunk is provided to Node 1 who will then read
the data from Node 3 directly.
4. Node 1 sends data to requesting client.

ECS Overview and Architecture 21

 Architecture

Figure 10. Read data flow for all-flash architecture

Note: In the all-flash architecture like EXF900, each node can read data from other node directly,
other than the hard disk drive architecture that each node can only read the data store in
themselves.

Write optimizations for file size

For smaller writes to storage ECS uses a method called box-carting to minimize impact to
performance. Box-carting aggregates multiple smaller writes of 2MB or less in memory
and writes them in a single disk operation. Box-carting limits the number of roundtrips to
disk required process individual writes.

For writes of larger objects, nodes within ECS can process write requests for the same
object simultaneously and take advantage simultaneous writes across multiple spindles in
the ECS cluster. Thus, ECS can ingest and store small and large objects efficiently.

Space reclamation
Writing chunks in an append-only manner means that data is added or updated by first
keeping the original written data in place and secondly by creating net new chunk
segments which may or may not be included in the chunk container of the original object.
The benefit of append-only data modification is an active/active data access model which
is not hindered by file-locking issues of traditional filesystems. This being the case, as
objects are updated or deleted, data in chunks becomes no longer referenced or needed.
Two garbage collection methods used by ECS to reclaim space from discarded full
chunks, or chunks containing a mixture of deleted and non-deleted object fragments
which are no longer referenced, are:
• Normal Garbage Collection - When an entire chunk is garbage, reclaim space.

22 ECS Overview and Architecture

 Architecture

• Partial Garbage Collection by Merge - When a chunk is 2/3 garbage, reclaim

the chunk by merging the valid parts of with other partially filled chunks to a new
chunk, reclaim space.
Garbage collection has also been applied to the ECS CAS data services access API to
clean up orphan blobs. Orphan blobs, which are unreferenced blobs identified in the CAS
data stored on ECS, will be eligible for space reclamation using normal garbage collection
methods.

SSD metadata caching

ECS metadata is stored in B-trees. Each B-tree may have entries in memory, journal
transactions and on disk. For the system to have a complete picture of a particular B-tree
all three locations are queried which often includes multiple look ups to disk.

To minimize latency for metadata lookups, an optional SSD-based cache mechanism has
been implemented in ECS 3.5. The cache holds recently accessed B-tree pages. This
means read operations on the latest B-trees will always hit the SSD-based cache and
avoids trips to spinning disks.

Here are some highlights for the new SSD metadata caching feature:
• Improved system-wide read latency and TPS (Transactions Per Second) for
small files
• One 960GB flash drive per node
• Net new nodes from manufacturing include the SSD drive as an option
• Existing field nodes—Gen3 and Gen2—can be upgraded using upgrade kits
and self-service installation
• SSD drives can be added while ECS is online
• Improvement for small file analytics workloads which require fast reads of large
data sets
• All nodes in a VDC must have SSDs to enable this feature
The ECS fabric detects when an SSD kit has been installed. This triggers the system to
automatically initialize and begin using the new drive. The following figure shows SSD
cache enabled:

ECS Overview and Architecture 23

 Architecture

Figure 11. SSD cache enabled

SSD metadata caching improves small reads and bucket listing. As we tested in our lab,
the listing performance improves 50% with 10MB objects. The read performance
improves 35% with 10KB objects and 70% with 100KB objects.

Cloud DVR
ECS supports cloud Digital Video Recording (DVR) feature which addresses a legal
copyright requirement for cable and satellite companies. The requirement is every unit of
recording mapped to an object on ECS needs to be copied a predetermined number of
times. The predetermined number of copies are known as fanout. The pre-determined
number of copies (fan-out) is not really a requirement for redundancy or performance
gain, but it is more of a legal copyright requirement for cable and satellite companies.
ECS supports:
• Create fanout number of copies of object created in ECS
• Allow read of specific copy
• Allow delete of specific copy
• Allow delete of all copies
• Allow copy of specific copy
• Allow listing of copies
• Allow bucket listing of fanout objects
The cloud DVR feature can be enabled through Service Console. For the first time, you
must enable the cloud DVR feature using Service Console. After enabling cloud DVR, by
default, for all the new nodes cloud DVR is enabled.

Run the following command in Service Console to enable the cloud DVR feature:
service-console run Enable_CloudDVR

24 ECS Overview and Architecture

 Architecture

The cloud DVR feature supports APIs and you can refer to the ECS Data Access Guide
for more details.

S3 select
S3 select, launched in version 3.7, enables applications to retrieve only a subset of data
from an object by using simple SQL expressions. By using s3 select to retrieve only the
data needed by your application, you can achieve drastic performance increases and
network bandwidth savings.

Using the example of a 2GB csv object, without s3 select an application would have to
download the entire 2GB object and then do the processing on that data. With s3 select,
the application issues SQL select commands and potentially gets only a small subset of
that data.

Figure 12. S3 select

S3 select can be used for objects in the csv, json, and parquet formats. It supports
querying gzip/bzip2 compressed objects of these three file types.

S3 select is commonly used by query engines, like presto. A connector in presto can
determine if a particular query can be sent directly to storage. For example, s3 select
pushdown.

AWS compliant S3 performs partial reads of an Object. This offloads query and sort to
ECS rather than using compute resources. This may provide a performance benefit for
use cases where network bandwidth and or compute resources are a bottleneck.

Note: S3 select is not enabled by default. It is suggested to enable it with 192GB of memory on
each node. Contact your Dell support team if you need to enable this feature.

Data movement
Data movement, also called copy-to-cloud, is a new feature in ECS 3.8.0.1 where a user
can copy local object data to an external S3 target, such as a secondary ECS that is not
federated, or to a public cloud target. (Currently, only AWS targets are supported).

ECS Overview and Architecture 25

 Architecture

Data movement is configured as a bucket option in the UI, as shown in the following
figure. It can be monitored by an account admin or system admin within the UI. The admin
can define policies about source and target buckets and criteria for objects. The admin
can also monitor the logs for all copy operations at the object level, including the copy
time, source object key, object size, target endpoint, duration, and result of the copy
operation (success/failure, error message). There are also alerts that show a summary of
all copy operations and errors on any failures.

Figure 13. Config data movement in a bucket

The data movement service can only run in Gen2 or later that have been upgraded to
192GB memory. It can only exist on IAM enabled buckets. Data movement policies
cannot sync deletes. This means that if an object is deleted from the source bucket, it will
not be deleted from the target bucket. The default scan interval (transfer frequency) is one
hour. The metadata search with LastModified index must be enabled in the bucket as
shown in the following figure.

26 ECS Overview and Architecture

 Architecture

Figure 14. Enable metadata search with LastModified

With a versioning enabled bucket, only the current version at the time of policy job
execution will be copied. File system (FS) enabled buckets are not supported because FS
buckets do not support IAM access.

We are extending our ecosystem to support a multi-cloud experience for Snowflake which
runs on public clouds in AWS. Dell and Snowflake customers can use on-premises data
stored on Dell ECS while keeping their data local or seamlessly copying it to public clouds
to leverage Snowflake’s ecosystem of cloud-based data analysis services.

The following workflow shows how Snowflake works with ECS Data movement:

1. An application writes data to an ECS local bucket.

2. A data movement policy in ECS is configured to copy all or a subset of the data to a
customer owned predefined staging bucket within AWS.
3. Data is written to the staging bucket.
4. This bucket will have S3 notifications set up to notify a customer owned AWS SQS
queue to which Snowflake is subscribed.
5. A Snowflake data pipeline process called Snowpipe wakes up and ingests the data
into Snowflake.
6. Data can then be deleted according to lifecycle policy in AWS.

ECS Overview and Architecture 27

 Architecture

Figure 15. Data movement with Snowflake

Fabric The Fabric layer provides clustering, system health, software management, configuration
management, upgrade capabilities and alerting. It is responsible for keeping services
running and managing resources such as disks, containers and the network. It tracks and
reacts to environment changes such as failure detection and provides alerts related to
system health. The Fabric layer has the following components:
• Node Agent - Manages host resources (disks, network, containers, and so on)
and system processes.
• Lifecyle Manager - Application lifecycle management which involves starting
services, recovery, notification, and failure detection.
• Persistence Manager - Coordinates and synchronizes the ECS distributed
environment.
• Registry - Docker image store for ECS software.
• Event Library - Holds the set of events occurring on the system.
• Hardware Manager - Provides status, event information and provisioning of the
hardware layer to higher level services. These services have been integrated to
support commodity hardware.

Node agent
The node agent is a lightweight agent written in Java that runs natively on all ECS nodes.
Its main duties include managing and controlling host resources (Docker containers,
disks, the firewall, the network) and monitoring system processes. Examples of
management include formatting and mounting disks, opening required ports, ensuring all
processes are running, and determining public and private network interfaces. It has an
event stream that provides ordered events to a lifecycle manager to indicate events
occurring on the system. A Fabric CLI is useful to diagnose issues and look at overall
system state.

Lifecycle manager
The lifecycle manager runs on a subset of three or five nodes and manages the lifecycle
of applications running on nodes. Each lifecycle manager is responsible for tracking
several nodes. Its main goal is to manage the entire lifecycle of the ECS application from
boot to deployment, including failure detection, recovery, notification, and migration. It
looks at the node agent streams and drives the agent to handle the situation. When a
node is down, it responds to failures or inconsistencies in the state of the node by

28 ECS Overview and Architecture

 Architecture

restoring the system to a known good state. If a lifecycle manager instance is down,
another one takes its place.

Registry
The registry contains the ECS Docker images used during installation, upgrade, and node
replacement. A Docker container called fabric-registry runs on one node within the ECS
rack and holds the repository of ECS Docker images and information required for
installations and upgrades. Although the registry is available on one node at a time, all
Docker images are locally cached on every node, so any may serve the registry.

Event library
The event library is used within the Fabric layer to expose the lifecycle and node agent
event streams. Events generated by the system are persisted onto shared memory and
disk to provide historical information about the state and health of the ECS system. These
ordered event streams can be used to restore the system to a specific state by replaying
the ordered events stored. Some examples of events include node events such as
started, stopped, or degraded.

Hardware manager
The hardware manager is integrated to the Fabric Agent to support industry standard
hardware. Its main purpose is to provide hardware specific status and event information,
and provisioning of the hardware layer to higher level services within ECS.

Infrastructure ECS appliance nodes currently run SUSE Linux Enterprise Server 12 for the
infrastructure. For ECS software deployed on custom industry standard hardware the
operating system can also be RedHat Enterprise Linux or CoreOS. Custom deployments
are done using a formal request and validation process. Docker is installed on the
infrastructure to deploy the encapsulated ECS layers. ECS software is written in Java so
the Java Virtual Machine is installed as part of the infrastructure.

Docker
ECS runs on top of the operating system as a Java application and is encapsulated within
several Docker containers. The containers are isolated but share the underlying operating
system resources and hardware. Some parts of ECS software run on all nodes and some
run on one or some nodes. The components running within a Docker container include:
• object-main - Contains the resources and processes relating to the data
services, storage engine, and the portal and provisioning services. It runs on
every node in ECS.
• fabric-lifecycle - Contains the processes, information, and resources required
for system-level monitoring, configuration management, and health
management. An odd number of fabric-lifecycle instances will always be
running. For example, there will be three instances running on a four-node
system and five instances for an eight-node system. fabric-zookeeper -
Centralized service for coordinating and synchronizing distributed processes,
configuration information, groups, and naming services. It is referred to as the
persistence manager and runs on odd number of nodes, for instance, five in an
eight-node system.

ECS Overview and Architecture 29

 Appliance hardware models

• fabric-registry - Registry of the ECS Docker images. Only one instance runs
per ECS rack.
There are other processes and tools that run outside of a Docker container namely the
Fabric node agent and hardware abstraction layer tools. The following figure provides an
example of how ECS containers can be run on an eight-node deployment:

Figure 16. Docker containers and agents on eight node deployment example

The following figure shows command line output of the docker ps command on a node
which shows the four containers used by ECS inside Docker. A listing is shown with all
the object-related services available on the system.

Figure 17. Processes, resources, tools, and binaries in object-main container

Appliance hardware models

Introduction

30 ECS Overview and Architecture

 Appliance hardware models

Flexible entry points enable ECS to rapidly scale to petabytes and exabytes of data. With
minimal business impact, an ECS solution can scale linearly in both capacity and
performance by adding nodes and disks.

ECS appliance hardware models are characterized by hardware generation. The third-
generation appliance series, known as the Gen3 or EX-Series, include three hardware
models. A high-level overview of the EX-Series is provided in this section. For complete
details, see the ECS EX-Series Hardware Guide.

Information about the first and second generation ECS appliance hardware is available in
the Dell ECS D- and U-Series Hardware Guide.

EX series EX series appliance models are based on standard Dell servers and switches. The
offerings in the series are:
• EX500 - The EX500 is the latest edition appliance which aims to provide
economy with density. With options for 12 or 24 drives, 2TB, 4TB, 8TB, 12TB,
16TB, and 20TB disk options (all same in the node). Clusters range from 120TB
to 7.68PB per rack. This series provides a versatile option for midsize
enterprises looking to support modern application and/or deep archive use
cases.
• EX5000 – Is the next generation of ECS dense appliance that refreshes the
EX3000. The EX5000 has a maximum capacity of 14 PB of raw storage per
rack using the 16TB and 20TB disks and can grow into exabytes across several
sites, providing a deep and scalable solution. The EX5000 is a 5U system that
can hold up to 100 drives per system. These nodes are available in two different
configurations known as EX5000S and EX5000D. The EX5000S is a single-
node chassis with options for 25, 50, 75 and 100 drives, and the EX5000D is a
dual-node chassis with 50 and 100 drives. The EX5000 can be used to expand
an existing EX3000 VDC, however, it requires a new rack due to higher power
needs. To add EX5000 to an existing rack, have your sales account team
submit a Request for Product Qualification (RPQ).
• EXF900 – The EXF900 is an all flash object storage solution of hyper-
converged nodes for low latency and high IOPs ECS deployments. With options
for 12 or 24 drives, 3.84TB, 7.68TB, and 15.36TB NVMe SSD drive options.
This platform starts at 230TB RAW minimum configuration and scales to
5.89PB RAW per rack. The following figure shows a node of EXF900.

Note: SSD Read Cache feature does not apply for EXF900; Cloud DVR is not supported on
EXF900; Tech Refresh is not supported with EXF900.

The EX-Series starting capacity options allow customers to begin an ECS deployment
with only the capacity needed, and to easily grow as needs change in the future. Refer to
the ECS Appliance Specification Sheet for more details on the EX-Series appliances
which also details the previous Gen2 U- and D- series appliances.

Post deployment updates to EX-Series nodes are not supported. These include:
• Changing the CPU.
• Upgrading hard drive size.

ECS Overview and Architecture 31

 Appliance hardware models

Note: EX300 and EX3000 were in EOL (end of sales life) in July 2022.

Appliance Starting with the release of the EX-Series appliances, a redundant pair of dedicated back-
networking end management switches are used. By moving to new appliance switch gear, ECS is
now able to adopt a front- and back-end switching mode of configuration.

The EX500, EX5000, and EXF900 appliances use the Dell S5248F for the front-end pair
of switches and for the pair of back-end switches. EXF900 appliances use the S5232F for
the aggregation back-end switches. Customers have the option of using their own front-
end switches instead of the Dell switches.

S5248F – front-end public switches

Dell offers an optional HA pair of front-end 25 GbE S5248F switches for customer network
connection to the rack. It has two 200GbE (QSFP28-DD) virtual link trunking (VLT) cables
per HA pair. These switches are called the Hare and the Rabbit switches. The following
figure shows a visual representation of how ports are intended to be used to enable ECS
node traffic and customer uplink ports.

Figure 18. Front-end network switch port designation and usage

S5248F – back-end private switches

Dell provides two 25 GbE S5248F back-end switches with two 200GbE (QSFP28-DD)
VLT cables. These switches are referred to as the Hound and Fox switches. All iDRAC
cables from nodes and all front-end switch management cable connections route to the
Fox switch. The following figure provides a visual representation of how ports are
intended to be used to enable ECS management traffic and diagnostic ports. These port
allocations are standard across all implementations.

32 ECS Overview and Architecture

 Network separation

Figure 19. Back-end network switch port designation and usage

S5232 – aggregation switch

Dell provides two 100GbE S5232F back-end aggregation switches (AGG1 and AGG2)
with four 100GbE VLT cables. These switches are referred to as the Falcon and Eagle
switches. In the following figure, all labeled ports indicate the port designations. This
configuration allows you to connect to seven racks of EXF900 nodes.

Figure 20. Aggregation switch port designation and usage

For more information about the networking and cabling, see the ECS EX Series Hardware
Guide.

Network separation
ECS supports separating different types of network traffic for security and performance
isolation. The types of traffic that can be separated include:
• Management
• Replication

ECS Overview and Architecture 33

 Security

• Data
There is a mode of operation called the network separation mode. In this mode each node
can be configured at the operating system level with up to three IP addresses, or logical
networks, for each of the different types of traffic. This feature has been designed to
provide the flexibility of either creating three separate logical networks for management,
replication and data, or combining them to either create two logical networks, for instance
management and replication traffic is in one logical network and data traffic in another
logical network. A second logical data network for CAS-only traffic can be configured,
allowing separation of CAS traffic from other types of data traffic like S3.

ECS implementation of network separation requires each logical network traffic to be

associated with services and ports. For instance, the ECS portal services communicate
using ports 80 or 443, so these ports and services will be tied to the management logical
network. A second data network can be configured; however, it is for CAS traffic only. The
following table highlights the services fixed to a type of logical network. For a complete list
of services associated with ports, refer to the latest ECS Security Configuration Guide.

Table 5. Services to logical network mapping

Services Logical network Identifier

WebUI and API, SSH, DNS, Management public.mgmt

NTP, AD, SMTP

Client data Data public.data

CAS data/S3 data public.data2

Replication data Replication public.repl

Dell Secure Connect Based on network Dell public.data or public.mgmt

Gateway Secure Connect Gateway is
attached

Note: ECS 3.7 allows S3 data access on both data (default) and data2 networks (though S3 is not
enabled by default on ECS versions earlier than 3.7 on the data2 identifier).

Network separation is achievable logically using different IP addresses, virtually using

different VLANs or physically using different cables. The setrackinfo command is used to
configure IP addresses and VLANs. Switch-level or client-side VLAN configuration is the
customer’s responsibility. For physical network separation, customers need to submit a
Request for Product Qualification (RPQ) by contacting Dell Global Business Service. For
more information about network separation refer to the ECS Networking and Best
Practices white paper that provides a high-level view of network separation.

Security

Introduction ECS security is implemented at the administration, transport and data levels. User and
administrator authentication are achieved using Active Directory, LDAP methods,
Keystone or directly within the ECS portal. Data level security is done using HTTPS for
data in motion and/or server-side encryption for data-at-rest.

34 ECS Overview and Architecture

 Security

Authentication ECS supports Active Directory, LDAP, Keystone, and IAM authentication methods to
provide access to manage and configure ECS; however, limitations exist as shown in the
following table. For more information about security, see the latest ECS Security
Configuration Guide.

Table 6. Supported authentication methods

Authentication
Supported
method

Active Directory • AD group support for management users

• AD group support for object user self-provisioning methods using self-
service keys using API
• Multi-domain is supported

LDAP • Management users may individually authenticate using LDAP

• LDAP Groups are NOT supported for management users
• LDAP is supported for Object Users (Self-service keys using API)
• Multi-domain is supported.

Keystone • RBAC policies not yet supported.

• No support for un-scoped tokens
• No support of multiple Keystone Servers per ECS system

IAM • Delivers identity federation and single sign-on (SSO) using SAML 2.0
standards
• Available through the S3 protocol only

Data services Object access using RESTful APIs is secured over HTTPS (TLS v1.2). Incoming requests
authentication are authenticated using defined methods such as Hash-based Message Authentication
Code (HBAC), Kerberos, or token authentication. The following table presents the
different methods used for each protocol.

Table 7. Data services authentication

Protocol Authentication methods

Object S3 V2 (HMAC-SHA1), V4 (HMAC-SHA256)

Swift Token - Keystone v2 and v3 (scoped, UUID, PKI tokens),

SWAuth v1

Atmos HMAC-SHA1

CAS Secret Key PEA file

File NFS Kerberos, AUTH_SYS

Data-at-rest Compliance requirements often mandate the use of encryption to protect data written on
encryption disks. In ECS encryption can be enabled at the namespace and bucket levels. Key
(D@RE) features of ECS D@RE include:
• Native low touch encryption at rest - easily enabled, simple configuration

ECS Overview and Architecture 35

 Security

• CIPHERs (AES-256 CTR) used

• RSA Public Key Encryption with 2048bit length
• External Key Management (EKM) cluster-level support:
▪ Gemalto SafeNet KeySecure
▪ IBM Security Key Lifecycle Manager
▪ Thales CipherTrust Manager
• Key rotation
• S3 encryption semantics support using HTTP headers such as x-amz-server-
side-encryption
• FIPS 140-2 compliance with US government cryptographic security standards

Note: FIPS 140-2 mode enforces the use of approved-only algorithms within D@RE; FIPS 140-2
compliance is only for the D@RE module, not the entire ECS product.

Gemalto SafeNet KeySecure will end of life on Dec 31, 2023. Refer to this External
Communication for more details. ECS customers who are using KeySecure can migrate
to CipherTrust Manager by opening a ticket with support.

ECS uses a key hierarchy to encrypt and decrypt data. The native key manager stores a
private key common to all nodes to decrypt the primary key. With EKM configuration, the
primary key is provided by the EKM. EKM provided keys reside in memory only on ECS.
They are never stored in persistent storage within ECS.

In a geo-replicated environment, when a new ECS system joins an existing federation, the
primary key is extracted using the public-private key of the existing system and encrypted
using the new public-private key pair generated from the new system that joined the
federation. From this point on, the primary key is global and known to both systems within
the federation. When using EKM, all federated systems retrieve the primary key from the
key management system.

Key rotation
ECS supports changing encryption keys. This can be done periodically to limit the amount
of data protected by a specific set of Key Encryption Keys (KEK) or in response to a
potential leak or compromise. A Rotation KEK Record is used with other parent keys to
create virtual wrapping keys for protecting Data Encryption Keys (DEK) and namespace
KEKs.

Rotation keys are natively generated or supplied and maintained by an EKM. ECS uses
the current Rotation Key to create virtual wrapping keys to protect any DEK or KEK
regardless of whether key management is done natively or externally.

During writes, ECS wraps the randomly generated DEK using a virtual wrapping key
created using the bucket and active rotation key.

As part of the rotation of keys, ECS re-wraps all namespace KEK records with a new
virtual primary KEK created from new rotation key, the associated secret context and the
active primary key. This is done to protect access to data protected by the previous
rotation keys.

36 ECS Overview and Architecture

 Security

Using an EKM affects the read/write path for encrypted objects. Rotation of keys allows
for extra data protection by using virtual wrapping keys for DEKs and namespace KEKs.
The virtual wrapping keys are not persisted and are derived from two independent
hierarchies of persisted keys. With the use of EKM, then the rotation key is not stored in
ECS and adds further to the security of data. We mainly add new KEK records and
update active ids but never delete anything.

Additional points to consider regarding key rotation on ECS are:

• The process of rotating keys only changes the current rotation key. The existing
primary, namespace, and bucket keys do not change during the key rotation
process.
• Namespace or bucket level key rotation is not supported, however, the scope of
rotation is at cluster level, so all new system encrypted objects will be affected.
• Existing data is not re-encrypted because of rotating keys.
• ECS does not support the rotation of keys during outages.
▪ TSO during rotation: Key rotation task suspended until the system comes out of
TSO.
▪ PSO is in progress. ECS must come out of a PSO before key rotation is
enabled. If a PSO happens during rotation, the rotation will fail immediately.
• Bucket encryption is not required to do object encryption using S3.
• Indexed client object metadata used as a search key is not encrypted.
See the latest ECS Security Configuration Guide for further information about D@RE,
EKM, and key rotation.

ECS IAM ECS Identify and Access Management (IAM) enables you to control and secure access to
the ECS S3 resources. This functionality ensures that each access request to an ECS
resource is identified, authenticated, and authorized. ECS IAM allows admin to add users,
roles, and groups. Admin can also restrict the access by adding policies to the ECS IAM
entities.

Note: ECS IAM is for use with S3 only. It is not enabled for CAS or filesystem enabled
buckets.

ECS IAM consists of the following components

• Account Management - enables you to manage IAM identities within each
namespace such as users, groups, and roles
• Access Management - access is managed by creating policies and attaching
them to IAM identities or resources
• Identity Federation - identity is be established and authenticated by SAML
(Security Assertion Markup Language). After the identity is established you will
use the Secure Token Service to obtain temporary credentials that will be used
to access the resource

ECS Overview and Architecture 37

 Security

• Secure Token Service - enables you to request temporary credentials for cross
account access to resources and also for users who are authenticated using
SAML authentication from an enterprise identity provider or directory service
By using IAM, you can control who are authenticated and authorized to use ECS
resources by creating and managing:

• Users - IAM user represents a person or application in the namespace that can
interact with ECS resources
• Groups - IAM group is a collection of IAM users. Use groups to specify
permissions for a collection of IAM users
• Roles - IAM Role is an identity that could be assumed by anyone who requires
the role. A role is similar to a user, an identity with permission policies that
determine what the identity can and cannot do.
• Policies - IAM policy is a document in JSON format, which defines the
permissions for a role. Assign and attach policies to IAM Users, IAM Groups,
and IAM Roles.
• SAML provider- SAML is an open standard for exchanging authentication and
authorization data between an identity provider and a service provider. SAML
provider in ECS is used to establish trust between a SAML-compatible Identity
Provider (IdP) and ECS
Each ECS system is allotted with an ECS IAM account. This account supports multiple
namespaces and has related IAM entities that are defined in its namespace.

• Individual namespaces support in managing account using the ECS IAM entities
such as users, roles, and groups.
• Policies, permissions, Access Control List (ACL) that are associated with the
ECS IAM entities, and the ECS S3 resources support in managing access to the
ECS IAM features.
• ECS IAM supports cross-account access using Security Assertion Markup
Language (SAML) and roles.
• ECS IAM supports Amazon Web Services (AWS) Access Key to access IAM
and S3 in ECS.
See the latest ECS Security Guide for more information about ECS IAM.

Azure AD OBO Support

Today more customers are moving to Azure AD, and more apps are using OIDC (OpenID
Connect), so that they can talk to a service provider like ECS that supports SAML
(Security Assertion Markup Language). Apps in this environment are using an Azure AD
On Behalf Of (OBO) workflow to exchange their OIDC token for a SAML assertion. With
the support of this new workflow, our customers can integrate their S3 applications to
authenticate identity.

The “OAuth 2.0 On-Behalf-Of” flow (OBO) serves the use case where an application
invokes a service/web API, which in turn needs to call another service/web API. The idea
is to propagate the delegated user identity and permissions through the request chain. For
the middle-tier service to make authenticated requests to the downstream service, it

38 ECS Overview and Architecture

 Security

needs to secure an access token from the Microsoft identity platform, on behalf of the
user.

ECS IAM features for S3 work with SAML identity providers to handle authentication and
SAML Assertion generation. It provides the following for applications:

• An application can authenticate with an identity provider and if successful receive a SAML
Assertion.
• The application uses the SAML Assertion to make a call to the ECS Secure Token Service
(STS) API (AssumeRoleWithSAML Method) to retrieve a temporary set of credentials to allow
the caller to assume a role.
• The application then performs ECS API calls that the role allows using the temporary
credentials.

Note: In the SAML model there are two main roles for a participant: Identify Provider and Service
Provider. Based on the ECS IAM SAML design ECS acts as the Service Provider and Azure AD
acts as the Identity Provider and generates SAML Assertions.

The “on-behalf-of (OBO)” flow describes the scenario of a web API using an identity other
than its own to call another web API. Referred to as delegation in OAuth, the intent is to
pass a user's identity and permissions through the request chain. For the middle-tier
service to make authenticated requests to the downstream service, it needs to secure an
access token from the Microsoft identity platform.

Object tagging Object tagging allows categorization of objects by assigning tags to the individual objects.
A single object can have multiple tags that are associated with it, enabling
multidimensional categorization.

A tag could describe some sort of sensitive information like a health record, or you can tag
an object to a certain product that can be categorized as confidential. Tagging is a sub-
resource of an object that has a life cycle integrated with object operations. You can add
tags to new objects when you upload them or add tags to existing objects. It is acceptable
to use tags to label objects containing confidential data, such as personally identifiable
information (PII) or protected health information (PHI). The tags must not contain any
confidential information, as tags can be viewed without having the actual read permission
to an object.

Additional information about object tagging

This section provides information about object tagging in IAM, object tagging with bucket
policies, handling object tagging during TSO/PSO, and object tagging during object
lifecycle management. Here are additional considerations:

• Object tagging in IAM

The key function of object tagging as categorization system comes when it is
integrated with IAM policies. This allows admin to configure specific user
permissions. For example, admin can add a policy that allows everyone to access
objects with a specified tag or you can configure and grant permissions to users,
who can manage the tags on specific objects. The other key aspect with object
tagging is how and where the tags are persisted. This is important because it has a
direct impact on various aspects of the system.

ECS Overview and Architecture 39

 Security

• Object tagging with bucket policies

Object tagging allows you to categorize the objects, additionally tagging gets
integrated with various policies. Lifecycle management policy allows you to
configure at a bucket level. Earlier versions of ECS supports Expiration, Abort
Incomplete Uploads, and Deletion of Expired Object Tagging Delete Marker. The
filter could include multiple conditions including a tag-based condition. Each tag in
the filter condition must match the key and the value.

• Object tagging during TSO/PSO

Object tagging is another entry set in system metadata; no special handling is
required during TSO/PSO. There is a set limit on the number of tags that are
allowed to be associated with each object, size of system metadata along with
object tagging is well within the memory limits.

• Object tagging during object lifecycle management

Object tagging is part of system metadata and handled simultaneously with system
metadata handling, during lifecycle management. The Expiration Logic and
Lifecycle Delete Scanner requires to understand tag-based policies. Object tags
enable fine-grained object lifecycle management in which you can specify a tag-
based filter, in addition to a key name prefix, in a lifecycle rule.

See the latest ECS Security Configuration Guide for further information about ECS object
tagging.

Object lock Dell ECS object lock protects object versions from accidental or malicious deletion such
as a ransomware attack. It does this by allowing object versions to enter a Write Once
Read Many (WORM) state, where access is restricted based on attributes set on the
object version.

Object lock is designed to meet compliance requirements such as SEC 17a4(f), FINRA
Rule 4511(c), and CFTC Rule 17.

For the best application compatibility, object lock is compatible with the capabilities of
Amazon S3 object lock.

Object lock overview

Object lock prevents object version deletion during a user-defined retention period.
Immutable S3 objects are protected using object- or bucket-level configuration of WORM
and retention attributes. The retention policy is defined using the S3 API or bucket-level
defaults (also set using the S3 API). Objects are locked during the retention period, and
legal hold scenarios are also supported.

There are two lock types for object lock:

• Retention period -- Specifies a fixed period of time during which an object
version remains locked. During this period, your object version is WORM-
protected and cannot be overwritten or deleted.
• Legal hold -- Provides the same protection as a retention period, but it has no
expiration date. Instead, a legal hold remains in place until you explicitly remove
it. Legal holds are independent from retention periods.

40 ECS Overview and Architecture

 Security

There are two modes for the retention period:

• Governance mode -- users cannot overwrite or delete an object version or alter
its lock settings unless they have special permissions. With governance mode,
you protect objects against being deleted by most users, but you can still grant
some users permission to alter the retention settings or delete the object if
necessary. You can also use governance mode to test retention-period settings
before creating a compliance-mode retention period.
• Compliance mode -- a protected object version cannot be overwritten or deleted
by any user, including the root user in your account. When an object is locked in
compliance mode, its retention mode cannot be changed, and its retention
period cannot be shortened. Compliance mode helps ensure that an object
version cannot be overwritten or deleted during the retention period.
Object lock requires the use of versioned buckets, enabling object lock on a bucket
automatically enables versioning. Once object lock is enabled, it is not possible to disable
it or suspend versioning for the bucket. Object locks apply to individual object versions
only, different versions of a single object can have different retention modes and periods.

An object can still be deleted, but the version still exists and is locked. Retention period
can be placed on an object explicitly, or implicitly through a bucket default setting. Placing
a default retention setting on a bucket does not place any retention settings on objects
that already exist in the bucket. Changing a bucket's default retention period does not
change the existing retention period for any objects in that bucket. In compliance mode,
locks cannot be removed, decreased, or downgraded to governance mode. In governance
mode, lock can be removed, bypassed, and elevated to compliance mode with the
appropriate assigned privilege.

Object lock and lifecycle

Objects under lock are protected from lifecycle deletions.

Lifecycle logic is made difficult due to variety of behavior of different locks. From lifecycle
point of view there are locks without a date, locks with date that can be extended, and
locks with date that can be decreased.
• For compliance mode, the retain until date cannot be decreased, but can be
increased:
• For governance mode, the lock date can increase, decrease, or get removed.
• For legal hold, the lock is indefinite.
Object lock condition keys
Access control using IAM policies is an important part of the object lock functionality. The
s3:BypassGovernanceRetention permission is important because it is required to delete a
WORM-protected object in governance mode (it is not effective for compliance
mode). IAM policy conditions have been defined in the following table to support object
lock:

ECS Overview and Architecture 41

 Data integrity and protection

Table 8. Object lock condition keys

Condition key Description

s3:object-lock-legal-hold Enables enforcement of the specified object legal

hold status

s3:object-lock-mode Enables enforcement of the specified object

retention mode

s3:object-lock-retain-until-date Enables enforcement of a specific retain-until-date

s3:object-lock-remaining-retention-days Enables enforcement of an object relative to the

remaining retention days

Note: Object lock requires ECS ADO and FS (File System) to be disabled on buckets when using
ECS versions 3.6.2 and ECS 3.7.x. Object lock is only supported by S3 API, no UI workflows. It
only works with IAM, and not with legacy accounts.

ECS object lock enhancements for ADO (Access During Outage) have been added
starting from ECS 3.8.0.1. It now supports ADO RO (Read Only) by default. For RW
(Read Write) mode, ECS will continue to deny setting Object Lock on ADO buckets by
default. There are flags at the namespace and individual bucket level that allow the user
to agree that they understand the risk of losing locked versions during TSO but would still
like to allow this feature. For help with enabling the object lock RW in ADO, see the latest
ECS Data Access Guide or ask the Dell support team. After flags are set on a bucket,
they cannot be disabled.

Table 9. Object lock support matrix

ECS version Setting flags Non-ADO ADO RO ADO RW

3.6.2/3.7/Partial
cannot set flags
Upgrade Yes No No

Full Upgrade to set to not allowed

3.8.0.1 (by default) Yes Yes No

Full Upgrade to
set to allowed
3.8.0.1 Yes Yes Yes

See the latest ECS Data Access Guide for more information about ECS object lock.

Data integrity and protection

Overview For data integrity, ECS uses checksums. Checksums are created during write operations
and are stored with the data. On reads, checksums are calculated and compared with the
stored version. A background task scan verifies checksum information proactively.

42 ECS Overview and Architecture

 Data integrity and protection

For data protection, ECS uses triple mirroring for journal chunks and separate EC
schemes for repo (user repository data) and btree (B+ tree) chunks.

Erasure coding provides enhanced data protection from a disk, node, and rack failure in a
storage efficient fashion compared with conventional protection schemes. The ECS
storage engine implements the Reed Solomon error correction using two schemes:
• 12+4 (Default) - Chunk is broken into 12 data segments. Four coding (parity)
segments are created.
• 10+2 (Cold archive) - Chunk is broken into 10 data segments. Two coding
segments are created.
ECS requires a minimum of five nodes with using the default of 12+4, the resulting 16
segments are dispersed across nodes at the local site. The data and coding segments of
each chunk are equally distributed across nodes in the cluster. For example, with eight
nodes, each node has two segments (out of 16 total). The storage engine can reconstruct
a chunk from any 12 of the 16 segments.

ECS requires a minimum of six nodes for the cold archive option, in which a 10+2 scheme
is used instead of 12+4. EC stops when the number of nodes goes below the minimum
required for the EC scheme.

When a chunk is full or after a set period, it is sealed, parity is calculated, and the coding
segments are written to disks across the fault domain. Chunk data remains as a single
copy that consists of 16 segments (12 data, 4 code) dispersed throughout the cluster.
ECS only uses the code segments for chunk reconstruction when a failure occurs.

When the underlying infrastructure of a VDC changes at the node or rack level, the Fabric
layers detect the change and trigger a rebalance scanner as a background task. The
scanner calculates the best layout for EC segments across fault domains for each chunk
using the new topology. If the new layout provides better protection than the existing
layout, ECS re-distributes EC segments in a background task. This task has minimal
impact on system performance; however, there will be an increase in inter-node traffic
during rebalancing. Balancing of the logical table partitions onto the new nodes also
occurs and newly created journal and B+ tree chunks are evenly allocated on old and new
nodes going forward. Redistribution enhances local protection by leveraging all of the
resources within the infrastructure.

Note: It is recommended not to wait until the storage platform is completely full before adding
drives or nodes. A reasonable storage utilization threshold is 70% taking consideration daily ingest
rate and expected order, delivery and integration time of added drives/nodes.

Compliance To meet corporate and industry compliance requirements (SEC Rule 17a-4(f)) for storage
of data, ECS implemented the following:

• Platform hardening - Hardening addresses security vulnerabilities in ECS such

as platform lockdown to disable access to nodes or cluster, all non-essential
ports (for example, ftpd, sshd) are closed, full audit logging for sudo commands
and support for Dell Secure Connect Gateway to shut down remote access to
nodes.

ECS Overview and Architecture 43

 Deployment

• Compliance reporting - A system agent reports system’s compliance status

such as Good indicating compliance or Bad indicating non-compliance.
• Policy-based record retention and rules - Ability to limit changes to records
or data under retention using policies, time-period and rules.
• Advanced Retention Management (ARM) - To meet Centera compliance
requirements a set of retention rules were defined for CAS only.
▪ Event based retention - Enables retention periods that start when specified
event occurs.
▪ Litigation hold - Enables temporary deletion prevention of data subject to legal
action.
▪ Min/max governor - Per bucket setting for minimum and maximum default
retention period.
Compliance is enabled at the namespace level. Retention periods are configured at the
bucket level. Compliance requirements certify the platform, and it is because of this that;
the compliance feature is only available for ECS running on appliance hardware. For
information about enabling and configuring compliance in ECS, see the current ECS Data
Access Guide and the most recent ECS Administrator’s Guide.

Deployment
ECS can be deployed as a single or multiple site instance. The building blocks of an ECS
deployment include:
• Virtual Data Center (VDC) - A cluster, also referred to as a site or
geographically distinct region, made up of a set of ECS infrastructure managed
by a single fabric instance.
• Storage Pool (SP) - SPs can be thought of as a subset of nodes and their
associated storage belonging to a VDC. A node can belong to only one SP. EC
is set at the SP level with either a 12+4 or 10+2 scheme. A SP can be used as a
tool for physically separating data between clients or groups of clients accessing
storage on ECS.
• Replication Group (RG) - RGs define where SP content is protected and
locations from which data can be accessed. An RG with a single member site is
sometimes called a local RG. Data is always protected locally where it is written
against disk, node, and rack failures. RGs with two or more sites are often
called global RGs. Global RGs span up to 8 VDCs and protect against disk,
node, rack, and site failures. A VDC can belong to multiple RGs.
• Namespace - A namespace is conceptually the same as a tenant in ECS. A key
characteristic of a namespace is that users from one namespace cannot access
objects in another namespace.
• Buckets - Buckets are containers for objects created in a namespace and
sometimes considered a logical container for sub-tenants. In S3, containers are
called buckets and this term has been adopted by ECS. In Atmos, the
equivalent of a bucket is a subtenant; in Swift, the equivalent of a bucket is a
container, and for CAS, a bucket is a CAS pool. Buckets are global resources in

44 ECS Overview and Architecture

 Deployment

ECS. Each bucket is created in a namespace and each namespace is created

in an RG.
ECS leverages the following infrastructure systems:
• DNS - (required) Forward and reverse lookups required for each ECS node.
• NTP - (required) Network Time Protocol server.
• SMTP - (optional) Simple Mail Transfer Protocol Server for sending alerts and
reporting.
• DHCP - (optional) Required if assigning IP addresses using DHCP.
• Authentication Providers - (optional) ECS administrators can be authenticated
using Active Directory and LDAP groups. Object users can be authenticated
using Keystone. Authentication providers are not required for ECS. ECS has
local user management functionality built-in however do note that users created
locally are not replicated between VDCs.
• Load Balancer - (required if workflow dictates, otherwise optional) Client load
should be distributed across nodes to effectively use all resources available in
the system. If a dedicated load balancer appliance or service is needed to
manage the load across ECS nodes, it should be considered a requirement.
Developers writing applications using the ECS S3 SDK can take advantage of
its built-in load balancer functionality. Sophisticated load balancers may take
additional factors into account, such as a server's reported load, response
times, up/down status, number of active connections and geographic location.
The customer is responsible for managing client traffic and determining access
requirements. Regardless of method there are a few basic options that are
generally considered including manual IP allocation, DNS Round Robin, Client-
Side Load Balancing, Load Balancer Appliances and Geographic Load
Balancers. The following are brief descriptions of each of those methods:
▪ Manual IP allocation - IP addresses are manually distributed to applications.
This is generally not recommended as it may not distribute load or provide fault-
tolerance.
▪ DNS round-robin - A DNS entry is created that includes all node IP addresses.
Clients query DNS to resolve fully-qualified domain names for ECS services
and are answered with the IP addresses of a random node. This may provide
some pseudo-load balancing. This method may not provide fault-tolerance
because often manual intervention is used to remove IP addresses of failed
nodes from DNS. Time to live (TTL) issues may be encountered with this
method. Some DNS server implementations may cache DNS lookups for a
period such that clients connecting in a close timeframe may bind to the same
IP address, reducing the amount of load distribution to the data nodes. Using
DNS for distributing traffic in a round-robin fashion is not recommended.
▪ Load balancing - Load balancers are the most common approach to
distributing client load. Clients can send traffic to a load balancer which
receives and forwards it on to a healthy ECS node. Proactive health checks or
connection state are used to verify each node’s availability to service requests.
Unavailable nodes are removed from use until they pass a health check.
Offloading CPU-intensive SSL processing can be used to free up those
resources on ECS.

ECS Overview and Architecture 45

 Deployment

▪ Geographic load balancing - Geographic load balancing leverages DNS to

route lookups to an appliance like the Riverbed SteelApp, for example, which
use Geo-IP or another mechanism to determine the best site to route the client
to.

Single site During a single site, or single cluster initial deployment, nodes are first added into a SP.
deployment SPs are logical containers of physical nodes. SP configuration involves selecting the
required minimum number of available nodes and choosing the default 12+4 or cold
archive 10+2 EC scheme. Critical alert levels may be set during SP configuration initially
and in the future; however, EC schema cannot be changed after SP initialization. The first
SP created is designated as the system SP and is used to store system metadata. The
system SP cannot be deleted.

Clusters generally contain one or two SPs, as shown in the following figure—one for each
EC schema; however, if an organization requires physical separation of data, additional
SPs are used to implement boundaries.

Figure 21. VDC with two storage pools, each configured with different EC scheme

After the initialization of the first SP, a VDC can be created. VDC configuration involves
designating replication and management endpoints. Although system SP initialization is
required prior to VDC creation, VDC configuration does not assign SPs but rather the IP
addresses of nodes.

After a VDC is created, RGs are configured. RGs are global resources with configuration
involving designating at least one VDC, itself, in the single- or initial-site setup, along with
one of the VDC’s SPs. An RG with a single VDC member protects data locally at the disk,
node, and rack level. The next section expands on RGs to include multisite deployments.

Namespaces are global resources created and assigned to an RG. At the namespace
level retention policies, quotas, compliance, and namespace administrators are defined.
Access During Outage (ADO) can be configured at the namespace level which is covered
in the next section. Generally, it is at the namespace level where tenants are organized.
Tenants may be an application instance or team, user, business group, or any other
grouping that makes sense to the organization.

Buckets are global resources that can span multiple sites. Bucket creation involves
assigning it to a namespace and an RG. The bucket level is where ownership and file or
CAS access is enabled. The following figure shows one SP in a VDC with a namespace
containing two buckets:

46 ECS Overview and Architecture

 Deployment

Figure 22. Single site deployment example

Multisite A multisite deployment, also referred to as a federated environment or federated ECS,

deployment may span across up to eight VDCs. Data is replicated in ECS at the chunk level. Nodes
participating in an RG send their local data asynchronously to one or all other sites. Data
is encrypted using AES256 before it is sent across the WAN over HTTP. Key benefits
recognized when federating multiple VDCs are:
• Consolidation of multi-VDC management efforts into a single logical resource
• Site-level protection in addition to locally to the node-, disk- and rack-level
• Geographically distributed access to storage in an everywhere-active strongly-
consistent manner
This section on multisite deployment describes features specific to federated ECS such as:
• Data consistency - By default ECS provides a strongly-consistent storage
service.
• Replication Groups - Global containers used to designate protection and
access boundaries.
• Geo-caching - Optimization for remote-site access workflows in multisite
deployments.
• ADO - Client access behavior during temporary site outage (TSO).

Data consistency
ECS is a strongly consistent system that uses ownership to maintain an authoritative
version of each namespace, bucket, and object. Ownership is assigned to the VDC where
the namespace, bucket or object is created. For example, if a namespace, NS1, is
created at VDC1, VDC1 owns NS1 and responsible for maintaining the authoritative
version of buckets inside NS1. If a bucket, B1, is created at VDC2 inside NS1, VDC2
owns B1 and responsible for maintaining the authoritative version of the bucket contents
and each object’s owner VDC. Similarly, if an object, O1, is created inside B1 at VDC3,

ECS Overview and Architecture 47

 Deployment

VDC3 owns O1 and is responsible for maintaining the authoritative version of O1 and
associated metadata.

The resiliency of multisite data protection comes at the expense of increased storage
protection overhead and WAN bandwidth consumption. Index queries are required when
an object is accessed or updated from a site that does not own the object. Similarly index
lookups across the WAN are also required to retrieve information such as an authoritative
list of buckets in a namespace or objects in a bucket, owned by a remote site.

Understanding how ECS uses ownership to authoritatively track data at the namespace,
bucket, and object level helps administrators and application owners make decisions in
configuring their environment for access.

Active replication group

During RG creation a Replicate to All Sites setting is available which is either left off, by
default, or can be toggled on which enables this feature. Replicating data to all sites means
that data written individually to each VDC is replicated to all other RG member VDCs. For
example, a federated X-number-of-sites ECS instance with an active RG configured to
replicate data to all sites will result in X times protection overhead, or X * 1.33 (or 1.2 in cold
archive EC) total data protection overhead. Replicating to all sites may make sense
especially for smaller data sets where local access is important. Leaving this setting off
means that all data written to each VDC will be replicated to one other VDC. The primary
site, where on object is created, and the site storing the replicate copy, each protect the
data locally using the EC schema assigned to the local SP. That is, that only the original
data is replicated across the WAN and not any associated EC coding segments.

Data stored in an active RG is accessible to clients by any available RG member VDC.

The following figure shows an example of a federated ECS built using VDC1, VDC2, and
VDC3. Two RGs are shown, RG1 has a single member, VDC1, and RG2 has all three
VDCs as members. Three buckets are shown, B1, B2, and B3.

In this example:
• Clients accessing VDC1 have access to all buckets
• Clients accessing VDC2 and VDC3 have access only to buckets B2 and B3.

Figure 23. Bucket-level access by site with single site and multisite replication groups

48 ECS Overview and Architecture

 Deployment

Passive replication group

A passive RG has three member VDCs. Two of the VDCs are designated as active and
are accessible to clients. The third VDC is designated passive and used as a replication
target only. The passive site is used for recovery purposes only and does not allow for
direct client access. Benefits of geo-passive replication are:
• Decrease in storage protection overhead by increasing the potential for XOR
operations
• Administrator-level control of the location used for replicate-only storage
The following figure shows an example of a geo-passive configuration whereby VDC 1
and VDC 2 are primary (source) sites that both replicate their data (chunks) to the
replication target, VDC 3:

Figure 24. Client access and replication paths for geo-passive replication group

Multisite access to strongly-consistent data is accomplished using namespace, bucket,

and object ownership across RG member sites. Inter-site across-the-WAN index queries
are required when API access originates from a VDC that does not own the required
logical construct(s). WAN lookups are used to determine the authoritative version of data.
Thus, if an object created in Site 1 is read from Site 2, a WAN lookup is required to query
the object’s owner VDC, Site 1, to verify if the object’s data that has been replicated to
Site 2 is the latest version of the data. If Site 2 does not have the latest version, it fetches
the necessary data from Site1; otherwise, it uses the data previously replicated to it. This
is illustrated in the following figure:

ECS Overview and Architecture 49

 Deployment

Figure 25. Read request to non-owner VDC triggers WAN lookup to object-owner VDC

The following figure shows the data flow of writes in a geo-replicated environment in
which two sites are updating the same object. In this example, Site 1 initially created and
owns the object. The object has been erasure-coded and the related journal transactions
written to disk at Site 1. The data flow for an update to the object received at Site 2 is as
follows:
1. Site 2 first writes the data locally.
2. Site 2 synchronously updates the metadata (journal write) with the object owner,
Site 1, and waits for acknowledgment of metadata update from Site 1.
3. Site 1 acknowledges the metadata write to Site 2.
4. Site 2 acknowledges the write to the client.

Note: Site 2 asynchronously replicates the data Site 1, the object owner site, as usual. If the data
must be served from Site 1 before it is replicated to it from Site 2, Site 1 will retrieve the data
directly from Site 2.

50 ECS Overview and Architecture

 Deployment

Figure 26. Update of the same object data flow in geo-replicated environment

In both read and write scenarios in a geo-replicated environment, there is latency involved
in reading and updating the metadata and retrieving data from the object-owner site.

Note: Starting from ECS 3.4, you can remove a VDC from a replication group (RG) in a multi VDC
federation without affecting the VDC or other RGs associated with the VDC. Removing VDC from
RG no longer initiates PSO (permanent site outage). Removing a VDC from RG initiates recovery.

See the latest ECS Administrator Guide for further information about Replication Group.

Geo-caching remote data

ECS optimizes response times for accessing data stored on remote sites by locally
caching objects read across the WAN. This can be useful for multi-site access patterns
where data is often fetched from a remote, or non-owner site. Consider a geo-replicated
environment with three sites, VDC1, VDC2 and VDC3, where an object is written to VDC1
and the replicate copy of the object is stored at VDC2. In this scenario, to service a read
request received at VDC3, for the object created at VDC1 and replicated to VDC2, the
object data must be sent to VDC3 from either VDC1 or VDC2. Geo-caching frequently
accessed remote data helps reduce response times. A Least Recently Used algorithm is
used for caching. Geo-cache size is adjusted when hardware infrastructure such as disks,
nodes, and racks are added to a geo-replicated SP.

Behavior during site outage

Temporary site outage (TSO) generally refers to either a failure of WAN connectivity or of
an entire site such as during a natural disaster. ECS uses heartbeat mechanisms to
detect and handle temporary site failures. Client access and API-operation availability at
the namespace, bucket, and object levels during a TSO is governed the following ADO
options set at the namespace and bucket level:
• Off (default) - Strong consistency is maintained during a temporary outage.
• On - Eventually consistent access is allowed during a temporary site outage.

ECS Overview and Architecture 51

 Deployment

Data consistency during a TSO is implemented at the bucket level. Configuration is set at
the namespace level, which sets the default ADO setting in place for ADO during new
bucket creation. This can be overridden at new bucket creation. This means that TSO can
be configured for some buckets and not for others.

Access during outage (ADO) not enabled

By default, ADO is not enabled, and strong consistency is maintained. All client API
requests where authoritative namespace, bucket or object data is required but temporarily
unavailable will fail. Object operations to read, create, update, delete, and list buckets not
owned by an online site, will fail. Also, operations of create and edit of bucket, user, and
namespace will also fail.

As previously mentioned, the initial site owner of bucket, namespace and an object, is the
site where the resource was first created. During a TSO, certain operations may fail if the
site owner of resource is not accessible. Highlights of operations permitted or not
permitted during a temporary site outage include:
• Creation, deletion, and update of buckets, namespaces, object users,
authentication providers, RGs, and NFS user and group mappings are not
allowed from any site.
• Listing buckets within a namespace is allowed if the namespace owner site is
available.
NFS enables buckets that are owned by the inaccessible site are read-only.

ADO enabled
In an ADO-enabled bucket, during a TSO, the storage service provides eventually
consistent responses. In this scenario reads and optionally writes from a secondary (non-
owner) site are accepted and honored. Further, a write to a secondary site during a TSO
causes the secondary site to take ownership of the object. This allows each VDC to
continue to read and write objects from buckets in a shared namespace. Finally, the new
version of the object becomes the authoritative version of the object during post-TSO
reconciliation even if another application updates the object on the owner VDC.

Although many object operations continue during a network outage, certain operations are
not be permitted, such as creating new buckets, namespaces, or users. When network
connectivity between two VDCs is restored, the heartbeat mechanism automatically
detects connectivity, restores service and reconciles objects from the two VDCs. If the
same object is updated on both VDC A and VDC B, the copy on the non-owner VDC is
the authoritative copy. So, if an object that is owned by VDC B is updated on both VDC A
and VDC B during synchronization, the copy on VDC A will be the authoritative copy that
is kept, and the other copy will be un-referenced and available for space reclamation.

When more than two VDCs are part of an RG, and if network connectivity is interrupted
between one VDC and the other two, then write/update/ownership operations continue
just as they would with two VDCS; however, the process for responding to read requests
is more complex, as described below.

If an application requests an object that is owned by a VDC that is not reachable, ECS
sends the request to the VDC with the secondary copy of the object. However, the
secondary site copy might have been subject to a data contraction operation, which is an
XOR between two different data sets that produces a new data set. Therefore, the

52 ECS Overview and Architecture

 Deployment

secondary site VDC must first retrieve the chunks of the object included in the original
XOR operation and it must XOR those chunks with the recovery copy. This operation will
return the contents of the chunk originally stored on the failed VDC. The chunks from the
recovered object can then be reassembled and returned. When the chunks are
reconstructed, they are also cached so that the VDC can respond more quickly to
subsequent requests. Note reconstruction is time consuming. More VDCs in an RG imply
more chunks that must be retrieved from other VDCss, and hence reconstructing the
object takes longer.

If a disaster occurs, an entire VDC can become unrecoverable. ECS treats the
unrecoverable VDC as a temporary site failure. If the failure is permanent, the system
administrator must permanently failover the VDC from the federation to initiate fail over
processing, which initiates resynchronization and re-protection of the objects stored on
the failed VDC. The recovery tasks run as a background process. You can review the
recovery progress in the ECS Portal.

An additional bucket option is available for read-only (RO) ADO which ensures object
ownership never changes and removes the chance of conflicts otherwise caused by
object updates on both the failed and online sites during a temporary site outage. The
disadvantage of RO ADO is that during a temporary site outage no new objects can be
created and no existing objects in the bucket can be updated until after all sites are back
online. The RO ADO option is available during bucket creation only, it cannot be modified
afterwards. By default, this option is disabled.

Note: Starting in 3.8.1, there is an option to enable CAS read operations on the target system
which is similar to eventual consistency. Please refer to the Administration Guide for more details.

Failure tolerance ECS is designed to tolerate a range of equipment failure situations using a number of fault
domains. The range of failure conditions spans a varying scope including:
• Single hard drive failure in a single node
• Multiple hard drive failure in a single node
• Multiple nodes with single hard drive failure
• Multiple nodes with multiple hard drive failures
• Single node failure
• Multiple node failure
• Loss of communication to one replicated VDC
• Loss of one entire replicated VDC
In either a single site, dual-site, or geo-replicated configuration, the impact of the failure
depends on the quantity and type of components affected. However, at each level, ECS
provides mechanisms to defend against the impact of component failures. Many of these
mechanisms have already been discussed in this paper but are reviewed here and in the
following figure to show how they are applied to the solution. These include:
• Disk failure
▪ EC segments or replica copies from the same chunk are not stored on the
same disk

ECS Overview and Architecture 53

 Deployment

▪ Checksum calculation on write and read operations

▪ Background consistency checker re-verifying checksums
• Node failure
▪ Distribute segments or replica copies of a chunk equally across nodes in a VDC
▪ ECS Fabric keeps services running and manages resources such as disks and
network.
▪ Partition records and tables protected by partition ownership failover from node
to node.
• Rack failure within VDC
▪ Distribute segments of replica copies of a chunk equally across racks in a VDC.
▪ One fabric registry instance runs in each rack and can be restarted on any
other node in the same rack should the node fail.

Figure 27. Protection mechanisms at the disk, node, and rack levels

Note: For the rack aware, when adding new rack to the exist cluster, some of the data will be
moved to the new rack to balance the data across all the racks equally. However, the process
could take a long period of time to avoid having a performance impact on the system. If the
customer keeps writing aggressively and fills the first rack, then all the new writes will happen only
on the new rack.

The following table defines the type and number of component failures that each EC
scheme protects against per basic rack configuration. The table highlights the importance
of considering the impact of protective failure domains on overall data and service
availability in terms of number of nodes required at each EC scheme.

Table 10. Erasure code protection across failure domains

# chunk
# nodes in
EC scheme fragments EC data protected against…
VDC
per node

12+4 5 or less 4 • Loss of up to four disks or

Default • Loss of one node

54 ECS Overview and Architecture

 Deployment

# chunk
# nodes in
EC scheme fragments EC data protected against…
VDC
per node

6 or 7 3 • Loss of up to four disks or

• Loss of one node and one disk from a second node

8 or more 2 • Loss of up to four disks or

• Loss of two nodes or
• Loss of one node and two disks

15 one node • Loss of up to four disks or

with 2
• Loss of three nodes or
fragments
and other • Loss of two nodes and one disk or
nodes with 1 • Loss of one node and two disks
fragment.

16 or more 1 • Loss of four nodes or

• Loss of three nodes and disks from one additional node or
• Loss of two nodes and disks from up to two different nodes or
• Loss of one node and disks from up to three different nodes or
• Loss of four disks from four different nodes

10+2 11 or less 2 • Loss of up to two disks or

Cold Storage • Loss of one node

12 or more 1 • Loss of any number of disks from two different nodes or

• Loss of two nodes

Disk Beginning in ECS 3.5 customers can replace failed disks with Dell services using an
replacement intuitive ECS portal (Web UI) workflow. The feature provides:
automation • Do-it-yourself resolution to drive failures
• Accelerated time to break fix
• Operational flexibility and TCO savings
The maintenance page in the ECS portal provides admin visibility to all disks in each
node. When a drive fails the system automatically initiates recovery. All types of resources
on drive are recovered and when the drive is ready to be removed from the node, ECS
portal will display the replace button, as shown in the following figure:

ECS Overview and Architecture 55

 Storage protection overhead

Figure 28. Disk replacement automation

Note: Only one drive can be replaced at a time. This is to avoid replacement of the wrong drive.

Tech refresh Tech refresh is a Dell Professional Services directed engagement available beginning in
ECS 3.5 to non-disruptively remove older hardware nodes from ECS clusters using the
embedded software feature. It is an efficient and low resource consuming operation that
can be precisely throttled. This feature reduces the overhead previously associated with
decommissioning of ECS hardware.

Tech refresh includes three parts:

• Node extend: Adding Gen3 nodes to existing cluster
• Resource migration: Move all resources from existing nodes to Gen 3 nodes
• Node evacuation: Clean up old nodes and remove them from the cluster
Professional Services should be involved during tech refresh maintenance. See the latest
ECS Tech Refresh Guide for more information about Tech Refresh.

Storage protection overhead

Each VDC member in an RG is responsible for its own EC protection of data at the local
level. That is, data is replicated but not any related coding segments. Although EC is more
storage efficient than other forms of protection, such as full copy drive mirroring, it does
incur an inherent storage cost overhead at the local level. However, when it is required to
have secondary copies replicated offsite, and to have all sites have access to data when a
single site becomes unavailable, the storage costs become more extensive than when
using traditional site-to-site data copying protection methods. This is especially true when
unique data is distributed across three or more sites.

ECS provides a mechanism in which storage protection overhead efficiency can increase
as three or more sites are federated. In a two-VDC replicated environment ECS replicates

56 ECS Overview and Architecture

 Storage protection overhead

chunks from the primary, or owner VDC to a remote site to provide high availability and
resiliency. There is no way to evade the 100% cost of protection overhead of a full copy of
data in a two-site federated ECS deployment.

Now, consider three VDCs in a multi-site environment, VDC1, VDC2, and VDC3, where
each VDC has unique data replicated to it from each of the other VDCs. VDC2 and VDC3
may send a copy of their data to VDC1 for protection. VDC1 would therefore have its own
original data, plus replicate data from VDC2 and VDC3. This means that VDC1 would be
storing 3X the amount of data written at its own site.

In this situation ECS can perform an XOR operation of VDC2 and VDC3 data locally
stored at VDC1. This mathematical operation compares equal quantities of unique data,
chunks, and renders a result in new chunk which contain enough characteristics of the
two original data chunks to make it possible to restore either of the two original sets. So,
where previously there were three unique sets of data chunks stored on VDC1,
consuming 3X the available capacity, there is now only two - the original local data set,
and the XOR reduced protection copies.

In this same scenario, if VDC3 becomes unavailable, ECS can reconstruct VDC3 data
chunks by using chunk copies recalled from VDC2 and the (C1 ⊕ C2) data from VDC3
stored locally at VDC1. This principle applies to all three sites participating in the RG and
is dependent upon each of the three VDC’s having unique data sets. The following figure
shows an XOR calculation with two sites replicating to a third site.

Figure 29. XOR data protection efficiency

If business service level agreements require optimum read access speeds even in the
event of a full site failure, then the replicate to all sites setting forces ECS to revert to full
copies of replicated data to be stored at all sites. Expectedly, this drives up the storage
costs in proportion to the number of VDCs participating in the RG. Therefore a 3-site
configuration would revert to 3X storage protection overhead. Replicate to All Sites setting
is available during RG creation and cannot be toggled back and forth.

ECS Overview and Architecture 57

 Conclusion

As the number of federated sites increases, the XOR optimization is more efficient in
reducing the storage protection overhead due to replication. The following table provides
information about the storage protection overhead based on the number of sites for
normal EC of 12+4 and cold archive EC of 10+2, illustrating how ECS can become more
storage efficient as more sites are linked.

Note: To lower replicated data overhead across three, and up to eight sites, unique data must be
written relatively equally at each site. By writing data in equal amounts across sites, each site will
have a similar number of replica chunks. Similar numbers of replica chunks at each site lead to
similar number of XOR operations that can occur at each site. Maximum multisite storage
efficiency is gained by reducing the maximum number of replica chunks stored by using XOR.

Table 11. Storage protection overhead

# sites in RG 12+4 EC 10+2 EC

1 1.33 1.2

2 2.67 2.4

3 2.00 1.8

4 1.77 1.6

5 1.67 1.5

6 1.60 1.44

7 1.55 1.40

8 (Max # sites in RG) 1.52 1.37

Conclusion
Organizations are facing ever-increasing amounts of data and storage costs, particularly
in the public cloud space. The ECS scale-out and geo-distributed architecture delivers an
on-premises cloud platform that scales to exabytes of data with a Total Cost of Ownership
that is significantly less than public cloud storage. ECS is a great solution because of its
versatility, hyper-scalability, powerful features, and use of commodity hardware.

58 ECS Overview and Architecture

 Resources

Resources
Dell.com/support is focused on meeting customer needs with proven services and
support.

Storage technical documents and videos provide expertise that helps to ensure customer
success on Dell storage platforms.

ECS Overview and Architecture 59