Business Case for a Common NFV Platform

Executive Summary
At its heart, Network Functions Virtualization (NFV) is based on the deployment of virtualized infrastructure. It is
the manner in which virtualization is deployed that will have the most significant impact on business outcomes.
Virtualization of network functions does not in itself guarantee the cost and agility promises of NFV.

This whitepaper, sponsored by VMWare and Affirmed Networks, considers two primary models of NFV
deployment: custom (software) stacks and common (virtualization) platform. By modeling the two virtualization
approaches against a traditional appliance-based architecture, the paper asserts that both virtualization
approaches have short-term benefits over the status quo but that the common platform approach is uniquely
able to support a sustainable business model whereas the traditional appliance-based and NFV custom stacks
approaches are likely to fail.

 Custom stacks TCO tracks requirements

growth
 Common platform TCO increases at half
the rate of requirements increase
 62% lower TCO than traditional appliance
approach
 33% one-year and >350% five-year ROI for
phased deployment

 Operator has total control—vendors

integrate to operator’s platform
 VNF vendors compete for operator’s
business
 Best-in-class deployments and control of
price points
Introduction
Network operators are facing an uncertain future with over-the-top (OTT) providers competing for
customers’ wallet share and the rising cost of infrastructure fueled by high rates of traffic growth, which
show no sign of declining (See Table 1).

Five-Year CAGR
Transactions
Network Type Throughput Connections
per Second
Fixed 36% 6% 10%
Mobile 44% 7% 13%
Table 1 – Growth Rates of Functional Requirements Drivers for Broadband Networks1

Traditional network architectures with their long deployment times and the resulting complex manual
and proprietary systems interfaces required to support them have been identified as the root causes of
high-cost, poor capacity scaling, and long innovation cycles. Operators are required to make large and
expensive capacity additions to their infrastructure well in advance of demand and on a service-by-
service basis (See Figure 1).

Figure 1 – Traditional Capacity Additions

In addition to investment inefficiencies, traditional appliance-based solutions are unable to react quickly
to new and changing service opportunities and needs, limiting incremental revenue opportunity and
increasing customer churn. Also, the need to deploy multiple technology silos to deliver a mixed
portfolio of services, together with the operational implications of multiple specialist teams, adds
further operational expense. Traditional architectures, consequently, are not capable of delivering a
sustainable business model.

1
See http://acgcc.com/wp-content/uploads/2015/03/Forecast-of-Mobile-Broadband-Bandwidth-
Requirements_ACG.pdf and http://acgcc.com/wp-content/uploads/2014/12/Forecast-of-Residential-Fixed-
Broadband-Requirements-2014.pdf.
2
To address these challenges operators are turning to Network Functions Virtualization (NFV)2 as an
enabler of new services, short service innovation cycles, and as a means to radically reduce the
operational cost of new and existing services.

NFV is explicitly designed to reduce cost and increase network scalability and agility. Figure 2 shows that
the agility of NFV has the potential to supply capacity that closely tracks demand and does so across
multiple services.

Figure 2 – NFV Capacity Additions

The potential of NFV to improve service agility and reduce total cost of ownership (TCO), however,
requires a platform approach that allocates hardware, software, and human resources to meet the
requirements for all services from a large pool of generally available resources in an on-demand fashion.
These criteria are used to evaluate two NFV platform approaches:
1. Custom software stacks that aim to integrate as much of the model as possible into a single
solution by a vendor.
2. A modular approach based on the deployment of a common virtualization platform where
multiple VNFs and other NFV components are provided independently.

Each platform approach is evaluated by comparing its TCO to the TCO of the traditional (appliance-
based) approach where all approaches are serving identical functional requirements demand.

Custom Software Stacks

The custom software stacks approach is often considered as a first step in virtualizing network functions.
A particular network function, vEPC for example, is chosen and then a vendor is selected to create
custom software stacks that replace the existing network appliances. This approach makes it easy to get
started with network function virtualization because the existing organization has the authority,
responsibility, and experience to make the move from appliance-based functions to virtual functions.
Once the initial function has been virtualized, the organization will begin virtualizing other functions.
This approach leads to the deployment of multiple software stacks, each stack representing a different
primary vendor for a service.

2
Network Functions Virtualisation–Introductory White Paper https://portal.etsi.org/NFV/NFV_White_Paper.pdf.
3
Figure 3 shows a schematic of the custom software stacks approach supporting three functions.

Figure 3 – Custom Software Stacks

In the custom software stacks the NFV vendor has abstracted the VNFs to run over commodity software
and hardware. Standardized frameworks, such as OpenStack, are used to create an open software stack.
The approach also includes NFV Management and Operations (NFV M and O) services. The use of
virtualization and abstraction principles by the custom software stack and its use of APIs instead of
custom scripts and manual processes do much to improve agility and reduce cost as compared to the
network appliance approach.

Standardized frameworks, however, require the vendor to make many unique design choices to create
the vertical software stack. In order to deliver the full capability of a virtualized solution, less mature
implementations may require the integrated MANO functions as a necessary component of the
operation of the underlying NFVI. This prevents a modular approach. Also, the commodity hardware and
associated software is frequently implemented on racks dedicated to a specific VNF. Traditional
technology is upgraded to virtualized technology but relatively little modification is made in the business
processes; the location of the hardware (it may remain in central offices or POPs rather than be moved
to cloud data centers) and staff and their roles and responsibilities remain unchanged. The custom
stacks create functional silos where the network function, the software and hardware used to host the
function, and people are dedicated to serving each function. This is a source of vendor lock-in; it hinders
and restricts communications across the organization and limits the ability to pool assets and staff
resources across all network functions.

NFV Common Platform

The common horizontal platform approach goes beyond simply replacing traditional appliances with
virtualized technology. It creates a horizontal platform as part of a modular deployment that eliminates
the organizational and technology silos that are implicit in the custom software stacks approach. This
gives the network operator total control of the platform with vendors integrating to the network
operator’s platform rather than the historical approach of a network operator implementing vendor-
based service silos. This creates an environment where the network operator induces VNF vendors to
compete for the network operator’s business and produces best-in-class deployments and control of the
price points. Conversely, deploying vendor-specific complete solutions (custom stacks) blurs the
boundaries among the modules and locks out competition.

Figure 4 provides a schematic of the common platform.

4
 Figure 4 – NFV Common Platform

This approach provides a NFV common platform where the virtual network functions are served as
multitenants of the common platform and the virtualized infrastructure that uses a service-oriented
approach to serve the VNF tenants. A fundamental principle of the common platform approach is that
the NFVI, MANO, and VNFs are established as modules with sharp boundaries among them. This
eliminates the barriers to resource sharing that are built into the custom stacks and appliance-based
approaches. It also allows the network operator to move quickly to NFV without limiting choice. There is
short time to benefit, because each module is developed separately and the modules stay within the
principles of the NFV approach.

TCO Comparison of Network Function Deployment Approaches

The total cost of ownership of the custom stacks and common platform approaches are compared to
that of the traditional approach for a large-scale deployment of NFV control and data plane nodes for
five years. In the first year the traditional approach has 50 control nodes and 75 data nodes.

Node requirements are determined by the prescribed throughput, concurrent connections or

transactions per second needed by a specific network function and the type of service being offered. For
example, value-added video service function capacity is determined because video streams are
throughput dependent. Machine-to-machine services capacity requirements, however, are determined
by the number of connections because very large numbers of devices are active, but data payloads are
small and signaling transactions are infrequent.

Table 1 (See Introduction) tabulates typical five-year compound annual growth rate (CAGR) for the most
common drivers of network functional requirements for broadband services.

The number of fixed and mobile broadband connections is projected to continue growing at moderate
rates because both markets are mature. Throughput, however, is growing rapidly on both networks. The
shift from watching broadcast TV to unicast video streaming is causing this explosive increase in the
throughput requirement. Transactions per second per connection are rising in the mobile broadband
market as older technologies are upgraded to 4G because the 4G signaling protocol is much chattier
than its predecessors. Fixed broadband transactions per second per connection also are expanding
because network operators are developing value-added service offerings in an attempt to boost average
revenue per user. Value-added services require more signaling transactions than do basic services.

5
The modeling scenario projects NFV control and data nodes to grow at 25 percent CAGR. This represents
a composite of the growth rates for throughput, connections, and transactions per second.

Elements of Total Cost of Ownership

Capital (capex) and expense items associated with the NFV infrastructure are modeled as annual
recurring expense. This includes all physical hardware; software required for virtualization, management
and orchestration of the NFV infrastructure; vendor support/maintenance fees; energy; floor space; and
engineering, management, and maintenance expenses incurred by the network operator. The cost of
VNF software and all associated operation expense items is excluded from the analysis.

Hardware and software prices are based upon market (street) prices as estimated by ACG Research.

TCO Comparison Results

Figure 5 compares the five-year TCO of the three approaches to network function deployment, and
Table 2 tabulates the five-year percentage cost savings compared to the appliance approach for each
TCO category.

Figure 5 – Five-Year TCO Comparison

Five-Year Expense Savings Compared to Appliance Approach

Expense Category Custom Stacks Common Platform
Capital 19% 30%
Environmental 60% 66%
Engr, Mgmt & Mnt 71% 80%
Total TCO 53% 62%
Table 2 – Five-Year Expense Savings Compared to the Appliance Approach

The custom stacks and common platform approaches significantly reduce TCO as compared to the
appliance approach. The cost savings are especially large for the engineering, management, and
maintenance expense category. These savings are caused by the application of the abstraction and
automation capabilities of both the custom stacks and common platform approaches. Capex and
environmental expenses also are reduced because COTS hardware is more cost and energy efficient
than purpose-built hardware. The common platform significantly extends these saving through its
horizontal integration approach across all resource categories. It breaks down the silos created by the
custom stacks approach and, thus, enables more extensive pooling of human resources and network
function virtualization infrastructure.
6
Common Platform Breaks Linkage of Cost Growth to Service Growth
The common platform is the only approach that achieves a key goal of the NFV initiative, the goal of
breaking the linkage between cost growth and requirements growth. Figure 6 shows that the annual
TCO increases of the custom stacks approach track closely with that of the requirements increases; the
TCO of the common platform barely increases.

Figure 6 – Annual Increases of TCO of Approaches versus Requirements Increases

Each incremental increase in requirements drives a corresponding incremental increase in custom stacks
capex: there are no economics of scale because infrastructure is not shared across the stacks. Custom
stacks labor costs do not scale because staff is dedicated to each vertical stack (silo). In contrast, the
common platform enjoys strong economies of scale and scope because its resources are pooled and
made available to all VNFs.

Additional Benefits of the Common Platform Approach

The common platform approach has additional benefits and advantages when compared to the other
approaches:
 Continuous asset efficiency increases (7 percent CAGR): All available compute, storage, and
networking resources are shared across all network functions. This permits an increase in asset
utilization without compromising service level agreements. Resources are shared for activities
such as development testing, high availability, and disaster recovery services. The cost of
software vendor support services are reduced through use of common software throughout the
infrastructure. Energy and floor space costs are reduced because asset utilization is increased.

 Continuous labor productivity increases (10 percent CAGR): The high levels of automation and
common operating procedures across the NFV infrastructure break the 1:1 linkage between
increased systems under management and the number of full-time equivalent support staff
required to operate and manage the NFV infrastructure3. Related costs such as training also are
reduced through continuous productivity improvements.

 Decreased time-to-market and increased service agility: The cloud-based operational model
and the flexible deployment of multivendor services is the source of decreased time-to-market,
increased service agility, and operational savings. Further operational savings are realized
through the replacement of per-vendor hardware support contracts with uniform routine

3
One large data center operator reports that it manages and operates 50,000 virtual machines that host 5,000
applications with a staff of 20 employees.
7
 maintenance schedules. Open innovation is achieved through rapid onboarding and service
chaining of new service functions through a common, multivendor virtualized platform. Asset
efficiency is maximized through the deployment of commodity hardware with no over-
provisioning required: all services access a common pool of spare capacity.

 Increased network flexibility and responsiveness: The increased flexibility and responsiveness,
and lower cost delivered by the common platform also create more revenue opportunities and
increased revenue growth rates. The lower cost point of the common platform stimulates
service demand by allowing profitable operation at lower price points. Increased agility allows
network operators to make trial service introductions more quickly and with a lower go-to-
market cost. This permits more trial services to be introduced within a fixed marketing budget
while increasing the probability of identifying viable new service offerings. Increased agility also
enables creation of differentiated service offerings to meet the unique needs of niche markets
quickly and at low cost. Time-sensitive services such as bandwidth on demand, bandwidth
calendaring, and even short-term service sales are made feasible by network responsiveness.

Return on Investment (ROI) for Phased Transition to NFV Common Platform

An ROI analysis compares the investment in a phased transition from an appliance-based (traditional
architecture) to NFV using the common platform with the savings (avoided cost) produced by operating
on the low-cost common platform. The scale of operations is the same as for the TCO comparison
(previously) with the same 25 percent CAGR in traditional node requirements. Figure 7 shows how the
transition from traditional nodes to NFV nodes is phased over eight years.

Figure 7 – Replacement of Traditional Node Capacity with NFV Common Platform

NFV common platform nodes are deployed to accommodate a portion of total capacity growth
requirements in the early study years; in the later study years they are used to accommodate all
capacity growth and to retire some of the traditional nodes. Deployment of NFV control nodes begins
immediately, but deployment of NFV data nodes begins in the third study year.

Annual total cost of ownership includes annual depreciation for hardware and software capex and
annual operation expense. Transition costs are the costs of moving from the traditional nodes to the
NFV common platform:
 Migration cost: The cost of moving a workload from a traditional node to a NFV node. There are
three categories of migration cost:
8
 1. Active migration: A traditional node that is not end-of-life is replaced with a NFV node.
Active node migration is costly.
2. Passive migration: A function that is on a traditional node that is end-of-life is moved to a
NFV node. Passive node migration costs little and may even produce a net savings.
3. New function: A new function is implemented on a NFV node rather than a traditional node.

 Project cost: The cost of setting up and making annual improvements to the NFV infrastructure
common platform:
o The initial project cost to set up the NFV infrastructure including the cost to build the
infrastructure and setting up connections to other systems and EMSs.
o Annual cost to improve the NFV infrastructure including adding new functionality to the
infrastructure and connecting to more systems and EMSs.
o Annual license cost of connectors between the EMSs and the infrastructure.
o Annual staff reduction cost: The cost to reallocate or outplace staff that is no longer needed
to build, manage or operate the traditional nodes.

 Education cost: This is the cost of the initial training of staff that is new to the NFV environment.

Hardware and software prices and labor rates are identical to those used in the total cost of ownership
analysis previously presented.

ROI for Phased Transition to NFV Common Platform Results

Figure 8 shows the annual TCO of the appliance-based approach, the total TCO of the phased move to
the common platform, and the savings (avoided cost) produced by the move to the common platform.

Figure 8 – TCO Comparison of Appliance-Based Approach with Phased Move to the Common Platform

The figure shows that the common platform approach breaks the linkage between requirements growth
and cost increase. The linkage break is produced by the much lower cost of NFV nodes as compared to
appliance-based nodes, the annual asset efficiency gains of the common platform, and the annual labor
productivity gains of the common platform.

Breakeven Point in Less Than One Year for Phased Move to the Common Platform
Figure 9 shows the ROI of the phased move to the common platform.

9
 Figure 9 – Return on Investment of Phased Move to the Common Platform

The breakeven point for the move to the common platform is less than one year. ROI is 33 percent in
the first year and grows to more than 350 percent by the fifth year. The rapid and high returns produced
by the gradual move to the NFV common platform reduce the financial risks of the transition to NFV and
give network operators the freedom to transition at a pace that best fits their operational capabilities.

Conclusion
The common platform approach alone breaks the link between growth in network requirements and
cost. Though the custom stacks approach produces an immediate cost reduction compared to the
appliance approach it is fundamentally flawed in that its costs continue to track requirements growth.
Its cost reduction delays the collapse of the network operator’s business model, but a custom stacks
does not prevent it. The common platform approach creates an operating environment and NFV
infrastructure platform that supports a sustainable business model.

An ROI analysis demonstrates the rapid payback and high ROI of a phased move to the common
platform from the appliance-based approach. The low-cost NFV nodes of the common platform and
ability to produce sustained labor productivity and asset efficiency gains are shown to achieve payback
in under a year and produce more than 350 percent ROI over five years.

ACG Research recommends that network operators consider the modular deployment of the ETSI
framework (common platform) approach, ensure that each module (NFVI, MANO/VIM, and VNF) is truly
modular, and can be integrated within a multivendor environment. This approach, however, goes
beyond the simple change-out of a legacy technology for new virtualized technology. It requires
rethinking (redesign) of fundamental business processes and the organizations that support them.

About Affirmed Networks: Affirmed Networks is the leader in virtualized Mobile Networks with over 20 customers
and 40 trials underway. Provides a complete, consolidated vEPC solution that runs together on a single, virtual
hardware instance for better performance, scalability, and cost featuring fully virtualized instances of SGSN,
MME/SGSN to support 2G,3G,4G, LTE and VoLTE; GGSN, SGW, PGW; WiFi Access functions: ePDG, TWAG/TWAP;
Policy Control functions: PCRF, OCS, PCEF; and more.

About VMMare: VMware is a global leader in cloud infrastructure and business mobility. Built on VMware's
industry-leading virtualization technology, our solutions deliver a brave new model of IT that is fluid, instant and
more secure. Customers can innovate faster by rapidly developing, automatically delivering and more safely

10
consuming any application. With 2014 revenues of $6 billion, VMware has more than 500,000 customers and
75,000 partners. The company is headquartered in Silicon Valley with offices throughout the world and can be
found online at www.vmware.com.

About ACG Research: ACG Research is an analyst and consulting company that focuses in the networking and
telecom space. We offer comprehensive, high-quality, end-to-end business consulting and syndicated research
services. Copyright © 2015 ACG Research. www.acgcc.com.

11