10 Gigabit Ethernet : Enabling Storage Networking for Big Data

Whitepaper

Introduction

Big data requires big networks.

Explosive data growth is a reality and the trajectory is continuing to be strong. In order to accommodate and support this level of
intensification, more robust and powerful networks are becoming more important than ever before. Data generation and the
diversification of data use drive the adoption of more role-based storage solutions within the data center. These factors, coupled
with the transition to highly virtualized data center environments, affects how organizations buy and manage server, storage, and
network assets and are key drivers in what is propelling Big Data into an everyday reality. The outlook is Big Data in the Cloud.

Big Data is comprised of datasets that grow so large that they become cumbersome to work with using on-hand database
management tools. Difficulties include capture, storage, search, sharing, analytics, and visualizing. The growth trend continues
because of the benefits of working with larger and larger datasets which allows analysts to discover business trends and solve
problems. Though a moving target, current limits are on the order of terabytes, exabytes and zetabytes of data.

There are various horizontals where Big Data is applicable, such as meteorology, genomics, biological research, Internet search,
finance, and business informatics. Arista is a strong player in this arena and many times the incumbent in all of these horizontals
providing great differentiation. One current difficulty of working with Big Data is the use of relational databases and desktop
statistics/visualization packages which require massively parallel software running on tens, hundreds, or even thousands of servers.
This notion directly ties in to the role of networking and cloud storage, and to the advantages of the Arista’s offering in the form of
high density, non-blocking, highly scalable networks in a small footprint.

Data is everywhere, whether it is from users, applications, or machines and it’s growing exponentially with no vertical or industry
being spared. Due to this reality, IT organizations everywhere are forced to come to grips with storing, managing and extracting
value from every piece of it -– as inexpensively as possible. This begins the real race to cloud computing where the framework
needs the ability to process data increasingly in real-time and in greater orders of magnitude -– and at a fraction of what it would
typically cost.

Application Drivers of Storage Networking

There are many key application drivers related to storage networking. The ones that rise to the top of the list are in the Cloud
Computing, Hadoop Cluster, Storage Virtualization, High Performance Computing, and Rich Content/Video arenas. Furthermore, the
rise of new web-based application architectures in the data center, the increasing use of virtualization tools to consolidate servers,
and the utilization HPC in core mission-critical applications all drive the need for a high performance, low latency, and highly
available networks for Big Data.

The key players driving Big Data are predominantly Internet service providers, application service providers, and storage service
providers. These players are among those who need 10 Gigabit Ethernet (GbE) bandwidth at the core of the network to support
applications as well as for remote replication and disk-to-disk backup. Also guaranteed to benefit from 10 GbE are applications such
as video editing and rich content applications that consume or generate huge amounts of data in a short time.

Hadoop Clusters

Data analytics has become a key element of the business decision process over the last decade. Classic reporting on a dataset stored
in a database was sufficient until recently, but yesterday’s data gathering and mining techniques are no longer a match for the
amount of unstructured data and the time demands required to make it useful. The common limitations for such analysis are
compute and storage resources required to obtain the results in a timely manner. A network that is designed for Hadoop
applications, rather than standard enterprise applications, can make a big difference in the performance of the Hadoop cluster.
Arista Networks satisfies Hadoop cluster network requirements has been successfully deployed in many Hadoop environments. As
costs fall and companies think of new ways to correlate and analyze data, Big Data analytics will become more common. Businesses
will especially benefit given their low-cost ability to manage and analyze Big Data.

Hadoop is a very powerful distributed computational framework that can process a wide range of datasets, from a few gigabytes to
petabytes of structured and unstructured data. Use of Hadoop has quickly gained momentum, and Hadoop is now the preferred
platform for various scientific computational and business analytics. While availability of commodity Linux based servers makes it
feasible to build very large clusters, the network is often the bottleneck, resulting in congestion and less efficient use of the cluster.

Hadoop consists of two key services: reliable data storage using the Hadoop Distributed File System (HDFS) and high-performance
parallel data processing using MapReduce. MapReduce is essentially the engine that brings speed and agility to the Hadoop
platform. With MapReduce, developers can create programs that process massive amounts of unstructured data in parallel across a
distributed cluster of processors or stand-alone computers. Hadoop allows enterprises to easily explore this complex data using
custom analyses tailored to their information and questions. Big Data problem is not just all about size of the data; it is also about
performance and how fast the data can be processed.

Arista offers high performance 1/10 GbE non-blocking, ultra-low latency solutions that can scale from a few racks to some of the
largest Hadoop deployments. In addition, Multi-Chassis LAG (MLAG) offers true active/active uplink connectivity from each rack,
allowing the full bi-sectional bandwidth of the network to be utilized in a flat layer 2 network. Arista’s Extensible Operating System
can easily be integrated with Ganglia and Nagios. Lastly, Arista’s networking solutions offer a true flat-line growth when it comes to
price for server-interconnect bandwidth. These factors make Arista’s networking solutions ideal for any Hadoop deployment.

Cloud Computing

Cloud computing helps organizations store, manage, share, and analyze their Big Data in a reasonable and simple-to-use way.
Today’s cloud Infrastructure-as-a-Service (IaaS) providers, supported by on-demand analytics solution vendors, make Big Data
analytics very affordable. As location-independent computing entails shared servers providing resources, software, and data
to systems and devices on demand, cloud computing is a very strong use driver for high performance storage networking.
Widespread adoption of virtualization and utility computing has caused this natural evolution and has resulted in customers no
longer needing expertise or control over the technology infrastructure that supports them. However, in order for those customers to
achieve the best performance and service possible, there needs to be robust and high performance networking in place to support it
such as what Arista Networks provides with their entire product line of switching solutions.

Storage Virtualization

Virtualization is a requirement for cloud computing as virtual nodes only need a subset of the performance of modern CPUs. As
aggressive as this convergence to virtual storage has been, it has also come with some disruptions. For one, it has equated to an
over-provisioning of storage and network capacity. Furthermore, there has been an excess of Fibre Channel (FC) storage and
Ethernet data network facilities and the propagation of new storage network technologies such as iSCSI. iSCSI over Ethernet is an
essentially ideal environment for virtualization due to the fact that it allows administrators to move virtual servers among physical
machines without reconfiguring the zoning and logical unit number (LUN) masking in the storage network. Together, these problems
can rapidly eliminate any hopes for benefits as the scope of virtual server deployment expands unless the network can play a
thriving role in making this transformation a success. Arista’s 10 Gigabit Ethernet solutions alleviate these disruptions and ensure
the full benefits of storage virtualization are met and realized.

High Performance Computing (HPC)

High performance computing involves the use of supercomputers or clusters of powerful processors to solve computationally
intensive problems. In order for a large number of processors to work together, clusters require interconnects that support high-
bandwidth and very low latency communication. Arista Networks has the lowest latency switch in the market today which is used
for HPC cluster interconnects to process large amounts of data in such industries as meteorology, genomics,oil and gas, biological
research or any other type of business that needs to crunch large amounts of data.

Rich Content / Video

Arista Networks is a key player in delivering a fully redundant, cloud based content origination service that hosts all video and high
resolution photo content. Arista Networks provides the Ethernet network to interconnect the streaming servers which require high-
performance 1/10GbE switching with very large packet buffering. This is perfectly suited to support high-bandwidth video streaming
and storage interconnection while maintaining the highest levels of system availability with both device and system level resiliency
designed into the rich content / video infrastructure. Additionally, it is often the data that needs to be aggregated, and analyzed
for business intelligence as well as for post-production. As reliability and resilience are important, the Arista Extensible Operating
System [EOS] with In-Service Software Upgrades and Stateful Fault Repair prove to be invaluable in maintaining service level
agreements and overall uptime.

Storage Networking Options for Big Data

10 Gigabit Ethernet deployments are rapidly growing as price and performance targets are met and as new optics enable broader
deployments. Additionally, the aggregate growth of new applications continues to increase bandwidth requirements. Although 10
Gigabit Ethernet is the optimal network interface, its role is really within a much larger overall picture of a switching solution.
Successful 10 Gigabit Ethernet deployments must incorporate leading intelligent switching services such as high performance, low
latency, high availability, and enhanced manageability to provide the necessary support for new applications.

A wide range of storage solutions exist in the market today, utilizing various approaches and a wide range of technologies. 10
Gigabit Ethernet prevails as the mainstream technology for Cloud Storage with iSCSI based block storage and network attached
storage (NAS). With non-blocking throughput, record density, low latency, and leading total cost of ownership, Arista Networks
switches are ideal for cloud storage applications.

Depending on their access method, storage systems are categorized as Storage Area Networks (SAN) and Network Attached Storage
solutions (NAS). In SAN environments, storage devices, although remote, appear as locally attached to the client, and access to
storage is block-based. In contrast, in NAS environments, clients access files remotely using a network-based file system.

Interconnect Technology Mapping Storage Area Networks

Protocol
A Storage Area Network (SAN) is an architecture whereby servers access
remote disk blocks across a dedicated Interconnect. Most SANs use the SCSI
protocol to communicate between the servers and the disks. Various
Fibre Channel (FC) Fibre Channel interconnect technologies can be used, each of them requiring a specific
Protocol (FCP) SCSI mapping protocol as shown to the left.

A SAN is a specialized network that enables fast, reliable access among

servers and external or independent storage resources and is the answer to
the increasing amount of data that needs to be stored in an enterprise
TCP/IP over Ethernet iSCSI network environment. By implementing a SAN, users can offload storage
traffic from daily network operations while establishing a direct connection
between storage elements and servers. SAN interconnects tie storage
interfaces together into many network configurations and across large
distances. Interconnects also link SAN interfaces to SAN Fabrics.
Ethernet FCoE
Switched SCSI, FCS, and Switched SSA form the most common SAN fabrics.
With gateways, SANs can be extended across WAN networks as well.
Switches allow many advantages in building centralized, centrally managed,
Infiniband iSER consolidated storage repositories shared across a number of applications.
Building a SAN requires network technologies with high scalability,
performance, and reliability in order to marry the robustness and speed of a
traditional storage environment with the connectivity of a network.

Fibre Channel Protocol (FCP)

Today, the majority of SANs use FCP to map SCSI over a dedicated Fibre
Channel (FC). Enterprises deploying Fibre Channel deploy multiple
networks including the LAN network, which typically uses Ethernet
technology (Ethernet is a basic component of 85% of all networks
worldwide, and is one of the most ubiquitous network protocols in
existence), as well as the dedicated FC network.

iSCSI

iSCSI has gained traction and attention as data centers look to lower costs
for robust storage. iSCSI rides on 1/10Gigabit Ethernet transport,
alleviating the complexity of a separate traditional Fiber Channel SAN. As
an ideal solution for many small and medium enterprise organizations,
iSCSI relies on TCP/IP protocols, making it a natural communication for
private and public cloud communications.

The performance advantages of iSCSI are compelling. Storage arrays must

keep up with new multicore processors and stack software that are now
capable of generating million iSCSI IOPS. New iSCSI arrays with 1/10 Gbps
Ethernet controllers such as Dell Equal Logic’s 6100 arrays, EMC VNX
 storage line, and Netapp’sFAS 6000 appliances combined with Arista
Networks 7000 switches can offer non-blocking storage access.

Network Attached Storage (NAS)

An increasingly popular method for consolidating storage resources is

Network Attached Storage (NAS). A NAS appliance is a server which
has the purpose of supplying file-based data storage services to other
devices on the network. NAS is a remote file system I/O where the file
request is redirected over a network.

NAS is recognized for three principal benefits, which in combination

lower overall TCO:

o Storage consolidation
o Deployment simplicity
o Ease of management

NAS systems have evolved to support, via a standard Ethernet network,

the storage tiering, high performance and high availability that had
previously only been available in SANs. This combined with its TCO
advantages has made NAS an increasingly adopted solution in the
enterprise.

Fibre Channel over Ethernet (FCoE)

The FCoE protocol is essentially an encapsulation of FCP over Ethernet. FCoE enables enterprise customers accustomed to Fibre
Channel to run the Fibre Channel Protocol directly over their LAN Ethernet network, hence allowing them to consolidate their LAN
and Storage network over the same network infrastructure. FCoE is aimed at organizations interested in keeping a high-end FC SAN,
yet are interested in LAN and SAN convergence. FCoE is at an early stage, and needs special changes to standard Ethernet (such as
Priority Flow Control), which Arista switches support.

Storage drives 10 Gigabit Ethernet At Arista, we view storage as a natural driver for 10 GbE cloud networking. The 7000 Family is
optimized for datacenter solutions, providing key storage characteristics such as resilience, high throughput, large buffers for
handling loss-less 1/10GBE traffic, PFC/ DCBX support and predictable low latency. Arista has designed a compelling 1/10 Gigabit
Ethernet architecture for iSCSI and NAS storage, as well as transparent transport support for FCoE traffic.

For FCoE to be functional enough to have success in the real

Cloud Storage world several considerations must be met and problems
FCoE remedied. First off, storage traffic must be adequately
Networking separated from other traffic running on the LAN. Next, no
storage packets can be dropped, as the Fibre Channel Protocol is
notoriously slow at recovering from packet errors. Finally,
Supported by all Ethernet proper Interoperability must exist with HBAs and gateways from
Brocade and Cisco only
vendors
different vendors. No two FCOE implementations are alike and
interoperability is not guaranteed. Nevertheless, Arista switches
Brocade and Cisco do support FCOE standards and are capable of carrying FCoE
Interoperable across
implementations don’t
multiple vendor solutions
interoperate traffic today over 10GEe networks.

In summary, iSCSI and NAS are the optimal options for "cloud
FC traditionally expensive,
Leading cost/performance storage", as opposed to FC and FCoE, due to the industry
FCoE does not fare better
standard, multi-vendor support, and cost-effectiveness. The
table to the left shows the strengths of iSCSI and NAS and
Benefit from the cost savings No cost benefit to illustrates why they are the best choice.
of convergence convergence
Arista Networks Advantages & Differentiating Features

Arista Networks is the leader in building scalable high-performance and ultra-low latency networks for today’s data center and cloud
computing environments. Purpose-built
built hardware and the Arista Extensible Operating System provide a single binary system image
across all platforms, maximum system uptime, sstateful fault repair, in-service
service upgrades, and a fully accessible Linux shell. Arista
Ari 10
GbE switches are the perfect network solution for your most demanding workloads. With support for VMware Virtualization and
hundreds of Linux applications integrated into hardware platforms specifically designed to meet the stringent power and cooling
requirements of today’s most demanding data centers, Arista delivers the most energy efficient and best performing 10 GbE G
platforms.

Storage at its very core is simply data. The more

re data that can be moved from host to target the moreore efficiently the host will
process applications. Arista switches have proven to be the highest performance, and top rated 10GbE switches available.
available The
lower the latency between a storage write and the acknowledgement from the storage target, the more efficiently the host can
process data and applications. Arista 10GbE switches are among the lowest latency available, ranging from 500ns
5 for 24-ports to
1.2usec in the 1RU 48-port 10GbE switch and 4.5usec at 384
384-port density with the Arista 7500 Series.

ent speeds buffering is a very important consideration. Arista

If the storage and host are connected at different Aris has a line of switches
with extremely large buffers, over 100x what other vend
vendors deliver. This absorbss bursty reads/writes and does not drop frames
reducing retransmissions, again improving efficiency and application performance. Arista is committed to an open storage market
mark -
one that equally supports iSCSI, NAS, SAN, FC, and FCFCoE. Arista is partnering with top storage system
em providers such as EMC,
Dell/EqualLogic, NetApp, and Panasas and top-tier
tier NIC vendors such as QLogic and Emulex to ensure customer success in deploying
these mission critical applications.

Cloud Storage Requirements Arista's Design Solution

Highly scalable to Petabytes of capacity High density switches & Fully non-blocking
non designs

Dynamic Buffer Allocation, Large buffers and VOQ architecture, & DCB/PFC
Lossless despite unpredictable traffic patterns
to provide lossless transmission

Multi-Chassis
Chassis Link Aggregation (MLAG) & Highly modular, self-healing
self EOS
High Availability
operating system

Virtualized servers and unified network access 10GigE to the servers and to storage

Visibility into virtualized infrastructure VMTracer

Fabric auto-configuration Supported

Simplified management Zero Touch Provisioning, DCBX, & EOS extensibility

Exceptional cost/performance, low power consumption per port, high

ROI (CAPEX and OPEX)
Density / compact footprint, & efficient front-to-rear
front airflow
Adding to the simplicity of Arista and its solutions comes the ability to provision your networks without the manual intervention
typically associated with doing so. Arista gives customers the ability to have Zero Touch Provisioning [ZTP] - Network Automation
for Cloud Data Centers. Using standards based protocols (e.g. DHCP, T/FTP, HTTP) the network can be rapidly provisioned.
Advanced scripting capabilities allow the administrator to tailor boot configurations based on a variety of parameters, meeting the
needs of even the most complex data center deployments. Combined with other Arista features such as VMTracer's adaptive VLAN
configuration, data center managers can fully automate the bring-up of network elements and virtual servers. Arista is unique with
it’s 'hands-off' provisioning to enable and automate the Cloud Data Center.

Applications

Fully automated deployment of a Virtualized Data Center : Arista switches deployed with ZTP, are downloaded with VMTracer
configuration parameters. As VMware vSphere servers boot and instantiate virtual machines, VMTracer automatically configures the
VLANS for the VMs. No manual configurations are necessary and the switch has auto-booted and is ready for use.

Using templates with Puppet/Cobbler to configure switches in an HPC cluster : Extending the model of cluster server
provisioning, the HPC manager uses templates to quickly deploy additional racks of servers in the HPC cluster. Along with the server
templates, the manager adds a template for the Top of Rack (ToR) switches as well. Now as racks of servers are added to the
cluster, both network and server platforms are configured and integrated into the existing cluster.

Using scripting to install Linux extensions to EOS : A service provider writes a shell script that is downloaded to a switch with ZTP
at boot time. The shell script downloads an OpenVPN RPM, installs and configures it. Its last function is to send a message to the
NOC with its logs and configuration information. After the switch reboots, managers can securely access the switch from the remote
NOC for provisioning.

Capability Ethernet FCoE Infiniband

1Gbps / 10Gbps
Port Speed 10Gbps 8/16/32Gbps
40Gbps / 100 Gbps
(Future)
Layer 3 No No
Yes

Switch Latency ~1usec 3.5usec - 10's usecs

~ 500ns

Interoperability with
Ethernet-based apps Seamless Not interoperable today Requires Gateway

Manageability Same Tools as Traditional

Same Tools as Traditional Limited Operational Tools
LAN
LAN

IEEE 802.1Qbb/az/au Capable

Capable N/A
PFC, Enhanced
Transmission Selection, Yes (L2 & L3) Yes (L2 only)
QCN Fat Tree
Summary

Coinciding with the developments of multi-protocol

protocol SAN and NAS, Arista Networks has taken concrete steps in extending the value of
FC and 10 GbE storage network assets in fast and evolving storage data center environments. Arista Networks has made great
strides and considerable investments in developing strong storage infrastructure business p
partners
artners around the globe. These partners
focus on large enterprises with advanced data center and storage requirements. With this collection of network based storage
services offerings that bring advanced capabilities to new environments, Arista Networks is the best-in-breed
breed choice for 10 GbE
G
deployments.

Cloud Storage for Big Data :

iSCSI and NAS, as opposed to FC or FCoE

iSCSI and NAS is where most of the opportunity for 10GbE is today

Key driver of 10GbE in the data center today

Arista has strong partnerships with many cloud storage vendors such as BlueArc,
Coraid, Dell EqualLogic, EMC, Isilon (now EMC), NetApp, & Panasas

Arista-based
based Cloud Server / Storage Design
Design Benefits :

Highly scalable
Scale-out cloud storage

Lossless

Highly available
High density servers

Simple to configure
High density TOR 1/10GigE switches

Virtualized with 10GigE or 40GigE uplinks

Dramatically lowers ROI

High density 10GigE/40GigE core
10 GbE empowers companies to expand application capabilities, reduce time to
solve complex financial, cloud, and clustered applications problems, and quickly
respond to changing customer needs and market conditions. High bandwidth, low low- 1/10GigE to the servers
latency, and energy efficient are the key differentiators of Arista Networks
switch offerings,, which address all of these drivers. Arista's 7000 family of 10
Gigabit Ethernet switches are uniquely suited to address the needs of cloud 10GigE to storage arrays
storage, featuring non-blocking 10 Gbps Ethernet
thernet throughput on every port
with sub-microsecond
microsecond latency and the highest port density per RU on the
market for fixed and modular form factors. Arista
Arista’s switches offer highly Low over-subscription
over ratio (2-5:1)
resilient Extensible Operating System (EOS) software, featuring self
self-healing
and live-patching
patching capabilities, and are designed for adequate and dynamic
buffering to deal with speed mismatch or avoid dropped packet or loss.

rista’s product family is designed for storage data centers—implementing high-performance

In summary, Arista’s performance compute
comput clusters; and
enterprise networks—consolidating
consolidating disparate SAN solutions into a common, low
low-cost, and high-speed
speed 10 GbE,
G IP storage network
while revolutionizing the way Big Data is networked
networked. The time is now, and with Arista you will always have the highest performance
and best of breed solution for your Big Data needs.

Information in this document is provided in connection with Arista Networks products.

For more information, visit us athttp://www.aristanetworks.com, or contact us at sales@aristanetwo
sales@aristanetworks.com