The ROI of Application Delivery Controllers in Traditional and Virtualized Environments

F5 White Paper

The ROI of Application
Delivery Controllers in
Traditional and Virtualized
Environments
How modern offload technologies in Application Delivery Controllers can
drastically reduce expenses in traditional and virtualized architectures, with
a fast ROI.

By Lori MacVittie
Technical Marketing Manager, Application Services

By KJ (Ken) Salchow, Jr.
Manager, Technical Marketing

White Paper
The ROI of Application Delivery Controllers in Traditional and Virtualized Environments

Contents
Introduction 3

The Magic of Server Offload 3

SSL Termination and Offload 5

Compression Offload 7

TCP Offload 9

Cashing In 10

Virtualization and Consolidation 11

Conclusion 14

2

White Paper

Introduction
The concept of spending money to make money—often referred to as “investing”
outside of the technology industry—is something just about every marketing
campaign promises, but few deliver. The ROI calculations to prove how quickly an
investment will reap return often come with a lot of conditions. For instance, it’s
only valid on Tuesdays, under a full moon, and when applied to a specific version of
software deployed on a (now) obsolete piece of hardware.

But solutions that provide a quick ROI along with significant technological benefits
do exist. The trick is finding these solutions and proving that the ROI model is valid
for almost every case.

It’s not magic. It’s simple math. In the following pages we won’t show you how to
determine if there is a compelling ROI case for Application Delivery Controllers, but
how to determine how much of a compelling case there really is.

The Magic of Server Offload
Let’s say you are in charge of a rather large data center for a rapidly growing web
2.0 company. And let’s say your “rather large” data center has approximately 1,000
servers. What if someone told you that you could reduce that server number by 40
percent without decreasing performance or availability? And what if that person
told you that the solution capable of this magical feat would pay for itself in just 10
months? After you stopped laughing, you might want to hear more about the magic
fairy dust that was going to reduce server count without impacting the application,
so you could laugh some more.

Assume each server costs an average of US $2,500, consumes 150 watts of power
at an average cost of 10.6 cents per KwH1 , and costs the organization $2882
a year in administrative costs. As this paper will show, reducing the number of
servers from 1,000 to 600, while servicing the same number of users at the same
performance levels, results in a full return on a $200,000 investment in about 10
months. The savings that achieve this ROI come from the reduction in power and
management costs those 400 servers would have required. Future savings can be
calculated by reducing the projected growth in server count and applying the same
cost savings to those servers as well.

3

White Paper

So, how can you realize these benefits? It’s not magic or fairy dust; it’s a
technological concept called “server offload” that moves computationally intensive
(CPU and memory) processing that would normally be handled by servers to an
external platform. That external platform is commonly known as an Application
Delivery Controller (ADC).

An ADC, in addition to performing commoditized functions like load balancing
(which you probably already know about from scaling out your 1,000-server
application) is also capable of offloading a variety of other functions such as SSL
termination and compression. Both of these tasks are highly computationally
intensive and CPU-bound, but are generally implemented at the server level instead
of within the application code. This makes them ideal functions to offload to a
device more efficient at handling such tasks. In addition, an ADC can add efficiency
to the connections themselves—resulting in additional savings.

Whether you are looking to consolidate physical resources and create a virtualized
data center, or you’re sticking with a tried-and-true traditional architecture, the
ability to forestall additional capital expenditures through the implementation
of server offload techniques can only improve your financial efficiency—while
maintaining or even improving availability, capacity, and performance. And with the
anticipated growth of virtual machines per server, it is imperative to ensure that each
application deployed within a virtual machine is as efficient as possible. The more
concurrent users or transactions per second that can be processed, while limiting 65% of organizations using
resources used, the more you can ensure that performance and capacity will not F5 BIG-IP solutions reported a
suffer as the number of virtual machines per physical server increases. payback period of 18 months
or less.
You may not run a data center of 1,000 servers; on the other hand, the reality is
that typical enterprise servers actually cost a bit more than $2,500, use more than Source: Survey of 192 F5
150 watts of power (since this is a typical idle draw), and have administrative costs BIG-IP users
much higher than $288 a year. So, even with a moderate number of servers to
manage, you will realize an excellent return on your investment in an ADC with
server offload capability. A 2009 TechValidate survey indicates that a majority of
TVID: 4F3-02B-15B
customers see an ROI of 18 months or less on their investment in an F5 Application
Delivery Controller.

Here’s how they do it.

4

White Paper

SSL Termination and Offload Certificate management is
another factor to consider
SSL is the most ubiquitous security protocol used in conjunction with websites when calculating the ROI for
SSL acceleration. Using an ADC
today. Data from Netcraft states that in January 2008 nearly 2.5 million sites on the
to offload SSL from servers
Internet made use of SSL.3 SSL enables clients and servers to encrypt and decrypt means all associated certificates
the data they exchange, securing it from prying eyes and from manipulation while in are managed in one place,
transit over public networks. on one device. This simplifies
management, reducing operating
SSL, like most mathematically complex algorithms, is CPU-intensive, requiring a expenses for an even greater ROI.
And because SSL offload enables
lot of CPU resources to churn through the mathematical computations required
the organization to legally use
to encrypt and decrypt large chunks of data. Additionally, because these complex one certificate per application
computations are being executed on general-purpose CPUs, the process of regardless of how many physical
encrypting and decrypting data can be a significant detriment to application and or virtual servers are required to
serve it, the organization saves
system performance.
even more on the cost of SSL
certificates.
One of the ways technology has addressed the problem of performance and
resource consumption is through hardware acceleration. The use of specialized
hardware—designed solely to perform the mathematical computations required
of SSL operations—simultaneously reduces the resources required and increases
the performance of such operations. Most of this kind of specialized hardware is
found in offload devices such as load balancers and ADCs, like F5 BIG-IP® Local
Traffic Manager™ (LTM). BIG-IP LTM offloads SSL processing by acting as a proxy
for web and application servers. Because the offload device performs all of the
SSL processing, the web and application servers can dedicate their resources to
responding to application requests. (See Figure 1.)

Testing, empirical evidence, and conventional wisdom place the amount of CPU
resources required for SSL processing (without hardware acceleration) at 30
percent of a typical server’s resources. If you have one server running at 90 percent
utilization, offloading the SSL processing to an ADC or to a load balancer will reduce
that utilization to 60 percent. Similarly, if your application currently requires 10
servers to support 1,000 users, then offloading SSL to an intermediate device should
reduce the number of servers required to seven, or increase the number of users you
can support on those 10 servers to 1,300.

Gartner, Inc., Worldwide Server Forecast, 2002-2014, September 2009 estimates
an annual average growth in server purchases of 4.48 percent for 2009-2012.4

5

White Paper

Figure 1: SSL offload with F5 BIG-IP LTM

1. BIG-IP LTM handles all SSL negotiations with the client. BIG-IP LTM receives
the encrypted request and decrypts it, then chooses a server and forwards
the request in plain text.

2. The server handles the request normally and returns it to BIG-IP LTM.

3. BIG-IP LTM encrypts the response and returns it to the client.
“Our business was able to save
CapEx using F5 BIG-IP LTM to
Deploying an ADC, and taking advantage of SSL offload, results in saving more than offload SSL certificates and
$40,000 in power costs alone by simply turning off 30 percent of the servers. With SSL processing. This reduced
an additional reduction in operating expenses in excess of $85,000 from simplified the number of SSL certificates
server administration, our example data center achieves full ROI in just 13 months— needed across multiple web
for the entire cost of the ADC—based on the SSL offload capability alone. (See servers, and reduced the
Table 1.) Additionally, the data center’s growth rate is effectively reduced, as it no overhead of those same web
longer requires four servers per month to support growing application demand. This servers.”
reduces capital expenditures, as fewer hardware server purchases are required.
IT Manager
Size of Data Center Cost of ADC 5
Payback ROI Medium Enterprise Computer
Small (125 servers) $40,000 22 months Software Company
Medium (500 servers) $120,000 17 months
Large (1,000 servers) $200,000 14 months

Table 1. ROI for Application Delivery Controller using SSL offload TVID: 420-53F-8C5

Based on savings realized due to reduction in administrative costs of $288 per server
per year, reduction in power usage from unnecessary servers using 150W at a cost of
$0.106/KwH, $2,500 cost of acquisition per server, a 30% reduction in number of servers
necessary, and a 4.48% year-over-year growth rate.

Again, it is important to note that this ROI is for the entire ADC solution, not just the
acquisition of the SSL capabilities alone.

6

White Paper

Compression Offload
Compression is commonly enabled on web servers as a means to lower costs by
reducing bandwidth utilization. It is also used to improve application performance.
Compression, like SSL operations, is mathematically intensive and is typically CPU-
bound. When used to compress dynamic content for which local server-based
caching is not available, compression can consume 4 to 30 times the CPU resources
that a server serving the same content without compression would utilize. This is
true for both Microsoft IIS and Apache web servers. (See Table 2.)

It is important to note that the decrease in bandwidth is significant and is therefore
providing benefits despite the increase in CPU utilization. Compression typically
affords applications a 3:1 reduction in size and improves application response time
dramatically, especially over high latency or bandwidth-constrained connections.

Bandwidth CPU Utilization
File Size
Decrease Increase
IIS 7.0 10KB 55% 4x
50KB 67% 20x
100KB 64% 30x
Apache 2.2 10KB 55% 4x
50KB 65% 10x
100KB 63% 30x

Table 2: Effect on CPU utilization of compression for dynamic
web application content7

But the hit on the CPU is significant enough to impact the overall capacity and
performance of the application (and any other applications deployed on the same
server). The magic of compression may well be negated by the need to deploy
additional servers to compensate for the reduction in processing power available.

On the other hand, the magic of offload can eliminate that need by taking on
the task of providing compression for the web/application server. ADCs and load
balancers can apply compression to content and generally take advantage of
hardware-assisted compression, which has a higher compression ratio (4:1 instead of
3:1) and provides more bandwidth savings than enabling compression on the web/
application server (See Figure 2.)

Offloading the task of compressing content—particularly dynamic content—
achieves a slightly better compression ratio while freeing the web/application server
CPU resources that would have been used to perform the task. Based on available

7

White Paper

ADCs that can intelligently
apply compression to content
only when it would benefit
performance and resource
consumption provide further
efficiencies on the ADC itself.
No cycles or memory are wasted
on content and connections
that would not benefit from
compression.

Figure 2: Compression offload using F5 BIG-IP Application Delivery Controller

1. BIG-IP LTM receives a web request and checks the client’s bandwidth,
then chooses a server and forwards the request.

2. The server handles the request normally and returns it to BIG-IP LTM.

3. BIG-IP LTM takes into consideration the available bandwidth and the
type of content and determines whether it will be a performance plus or
negative to apply compression. It then acts on the decision and returns
the response to the client.

data regarding the impact of compression on CPU utilization and current average
sizes of web pages, we can assume that an average of 20 percent of a server’s
resources are consumed by the process of applying compression. When offloaded to
50% of IT organizations report
an external device such as an ADC, those resources can be refocused on the server’s
that they reduced annual OpEx
primary task of serving applications.
budgets by 10% to 20% or
As with SSL offload, this results in either a reduction in servers needed to support more by deploying F5 BIG-IP
capacity needs, or in an immediate increase in capacity. (See Table 3.) If 1,000 users solutions.
are being supported by 10 servers, offloading compression to an ADC should result
Source: Survey of 200
BIG-IP users

TVID: 5DD-99E-9B6

8

White Paper

in those same 1,000 users being supported by only eight servers, or in total user
capacity being increased to 1,200 using 10 servers.

Size of Data Center Cost of ADC5 Payback ROI
Small (125 servers) $40,000 34 months

Table 3. ROI for Application Delivery Controller using compression offload

Based on savings realized due to reduction in administrative costs of $288 per server/per
year, reduction of power from unnecessary servers using 150W at a cost of $0.106/KwH,
$2,500 cost of acquisition per server, a 20% reduction in number of servers necessary, and
a 4.48% year-over-year growth rate.

TCP Offload
The use of the term “TCP offload” to describe TCP multiplexing is somewhat of an
anomaly. Other offload technologies actually offload complete functionality from
servers, while TCP offload uses optimization of resources to offload TCP overhead
and dramatically increase capacity of servers.

TCP offload, more often referred to as TCP multiplexing, is an optimization technique
common to ADCs that exploits the nature of persistent connections to achieve
higher utilization of TCP connections by sharing them on the back end, across users.
Because the full-proxy architecture of ADCs requires two separate networking
stacks, user TCP connections are made to the intermediary (the ADC), while server
connections are maintained between the intermediary and the servers. This allows
the intermediary to maintain a much higher number of user connections than may “We’ve been able to leverage
actually be supported by the server infrastructure, effectively increasing the capacity the F5 (solution) in ways that
of the servers. we didn’t expect when we
purchased it.”
In any architecture, each client connection requires a matching connection on the
server. This usually means a range of two to six connections per user are consumed IT Architect,
on the server, whether virtual or traditional. Using TCP multiplexing, the user Global 500 Professional
still opens two to six connections to the “server” but the intermediary brokers Services Company
those connections and instead opens only one to the server and then reuses that
connection across the user session. The ADC will also reuse that same connection
TVID: 956-3C2-AD6

9

White Paper

for additional users, opening new connections to the server only when necessary to
maintain application availability and performance.

This magical “offload” has dramatic results: a reduction of 66 to 90 percent6 in
server-side connections with an improvement in performance as measured by Time
To First Byte (TTFB).

What that means in practical terms is that you can serve the same number of
concurrent users with one-third the physical hardware—or one-third the same
number of virtual instances of the application. We’ll keep our estimates for ROI
purposes on the low end, using a 66 percent reduction for calculations. (See Table 4.)


Table 4. ROI for Application Delivery Controller using TCP multiplexing

Based on savings realized due to reduction in administrative costs of $288 per server/per
year, reduction of power from unnecessary servers using 150W at a cost of $0.106/KwH,
$2500 cost of acquisition per server, a 66% reduction in number of servers necessary, and
a 4.48% year-over-year growth.

Cashing In
The individual value to your organization from any one of these offload technologies
can be significant, but putting them all together amplifies their value. If we take
the 30 percent reduction from SSL offload, apply the 20 percent compression
improvement, and apply a 66 percent reduction through TCP optimization, we
get a roughly 81 percent overall resource reduction. In this pristine example, our
1,000-server data center would see an ROI of its entire initial ADC deployment in five

10

White Paper

months—just from the offload technologies alone. (See Table 5.) This doesn’t even
begin to consider the increased uptime and operational savings. While compelling,
it’s reminiscent of that magic fairy dust. However, if we assume even half of that
overall reduction, still with very moderate power, server, and management costs, the
ROI is still quite compelling.


Table 5.ROI for Application Delivery Controller using combined offload technologies

Based on savings realized due to reduction in administrative costs of $288 per server/
per year, reduction of power from unnecessary servers using 150W at a cost of $0.106/
KwH, $2,500 cost of acquisition per server, and a 40% reduction in number of servers
necessary.

Virtualization and Consolidation
Many organizations look to server virtualization and data center consolidation to
achieve the same kinds of OpEx and CapEx savings we have already demonstrated
can be achieved using ADC offloading technologies. While the savings from
virtualization can be substantial, additional investment is required, not to mention
the challenges of maintaining operations as the organization moves from a one-
to-one physical server environment to a virtual one. The value of an ADC to a
“When virtualizing, set up a
virtualized infrastructure can be quite significant, even without using offload
planning scenario and acquire the
technology. latest advanced tools. Such tools
include capacity, placement and
However, it is important to note that while traditional ADC features (load balancing performance to automate and
and traffic management) enable a seamless transition between physical and virtual ‘metricize’ the planning process.
machines, ADC offload technologies are even more relevant in the virtualized Use total-cost-of-ownership
analytics to ascertain optimum
environment, where the dynamic nature of the environment makes managing SSL
virtual machine placement and to
certificates and TCP connection setups and teardowns more challenging. By centrally determine the benefits.”
managing SSL certificates and application-wide compression profiles, you can
alleviate the need to do this at the virtual machine (VM) level every time you spin up Gartner, Inc.
or spin down another server, further reducing the management costs of those VMs. Ten Helpful Hints for Reducing
Server Infrastructure Costs
Using an ADC, each VM has more processing power available for the application,
November 2008
effectively increasing its capacity and alleviating the need to deploy more VMs that
can constrain the architecture and increase management costs. This can drastically

11

White Paper

accelerate and amplify the ROI of your virtualization and consolidation efforts.

Let’s assume that the 1,000-server data center from our example implements a
100 percent virtualization effort, converting all existing physical servers into VMs
at a fairly respectable rate of 15 VMs to one physical machine. This requires some
significant investment in new hardware and licensing, but, over time, generates
significant savings. (See Table 6.)

Consolidate Savings
Year Hardware Software Mgmt. Power Total Cumulative
1 $ (378,000.00) $ (196,000.00) $ 126,138.36 $ (447,861.64) $ (447,861.64)

2 $ (16,200.00) $ (8,400.00) $ 131,852.91 $ 107,252.91 $ (340,608.73)

3 $ (21,600.00) $ (11,200.00) $ 137,567.47 $ 104,767.47 $ (235,841.26)
4 $ (16,200.00) $ (8,400.00) $ 143,839.54 $ 119,239.54 $ (116,601.72)
5 $ (16,200.00) $ (8,400.00) $ 150,390.37 $ 125,790.37 $ 9,188.65

Table 6. Virtualization savings

Based on a 15:1 physical machine reduction with new hardware costs of $5,400 per platform,
$2,800 in virtualization software per platform, using 300W per new platform at a cost
of $0.106/KwH, no reduction in the actual number of “servers” or VMs resulting in no
management savings, and a 4.48% year-over-year VM growth rate.

While these calculations are certainly not intended to provide guidance in the costs
and savings of real-world virtualization and consolidation efforts, and do not reflect
the myriad variables associated with such a drastic architectural change (like the
reduced cost of cooling due to reduced BTU output or the savings in not having to
build new facilities), they are sufficient to show the significant impact that adding
server offload technology to your virtualization efforts can provide.

Based on these calculations, it is well into the fifth year of operation before the
effort breaks even and the organization starts to see a positive ROI. After that point,
the organization begins to see a substantial savings year-over-year.

However, using the same scenario, but adding an ADC with server offload
technology to reduce the number of VMs by 40 percent, can accelerate the
organization’s ROI in two ways. (See Table 7.)

12

White Paper

Consolidate/Offload Savings
Year Hardware Software Mgmt. Power Total Cumulative
1 $ (426,800.00) $ (117,600.00) $ 120,384.00 $ 133,943.60 $ (290,072.40) $ (290,072.40)

2 $ (10,800.00) $ (5,600.00) $ 125,568.00 $ 139,936.92 $ 249,104.92 $ (40,967.48)

3 $ (10,800.00) $ (5,600.00) $ 131,328.00 $ 146,208.99 $ 261,136.99 $ 220,169.51
4 $ (10,800.00) $ (5,600.00) $ 137,088.00 $ 152,759.82 $ 273,447.82 $ 493,617.33
5 $ (10,800.00) $ (5,600.00) $ 143,424.00 $ 159,589.41 $ 286,613.41 $ 780,230.75

Table 7. Savings on virtualization using Application Delivery Controller
with offload technologies

Based on a 15:1 physical machine reduction with new hardware costing $5,400 per platform,
$2,800 in virtualization software per platform, using 300W per new platform at a cost of
$0.106/KwH, a 4.48% year-over-year VM growth rate, an initial $200,000 ADC investment,
and 40% reduction in VMs due to server offload technology.

First, notice that the break-even point is a full two years sooner (cumulative savings
become positive in year 3). Despite the additional upfront costs of the ADC, the
initial costs are about 50 percent lower as a result of the reduction in hardware and
VM licenses needed, lower VM management costs, and additional power savings.

Second, because that investment is significantly reduced on a year-over-year basis,
and there is an associated ongoing reduction in management costs (because the
number of new VMs grows more slowly), the positive ROI is more than doubled on
a yearly basis.

Again, the real costs and savings for any particular virtualization effort depend on
a great number of variables that are beyond the scope of this discussion. However,
the reality is plain to see. Whatever the actual costs and benefits of virtualization
efforts are, adding server offload technology can demonstrably lower those costs
and amplify those benefits.

13

White Paper

Conclusion
The compute resource cost of mathematically complex functions, such as SSL
operations and compression, is a significant burden on web and application servers.
These complex operations are CPU-bound and consume resources in a way that
negatively impacts application performance and capacity. This is true whether the
server in question is traditional or virtual. So, it’s not magic or pixie dust, nor is it a
trick with numbers: the ROI offered by an Application Delivery Controller employing
SSL offload, compression offload, and TCP optimization is real. In addition,
implementing these technologies in conjunction with existing virtualization and
consolidation efforts can amplify an organization’s cost savings and accelerate the
overall ROI.

1
Energy Information Administration: see www.eia.doe.gov/cneaf/electricity/epm/table5_3.html for averages across
industries 1995-2009; see www.eia.doe.gov/cneaf/electricity/epm/table5_6_b.html for specific industry and state
guidance. Note that figures for 2008 and 2009 are estimated.
2
Based on 1 hour/month/server at US $24.
3
Netcraft, http://news.netcraft.com/SSL-Survey/.
4
Gartner, Worldwide Server Forecast Database, September 15, 2009.
5
Assumes a high availability implementation comprising two ADCs.
6
F5 Deployment Guide, Tuning the OneConnect Feature on the BIG-IP Local Traffic Manager
7
Web Performance, Inc., Measuring the Performance Effects of Dynamic Compression in IIS 7.0
Web Performance, Inc., Measuring the Performance Effects of mod_deflate in Apache 2.2
Intel® Software Network, HTTP Compression for Web Applications

F5 Networks, Inc. 401 Elliott Avenue West, Seattle, WA 98119 888-882-4447 www.f5.com

F5 Networks, Inc. F5 Networks F5 Networks Ltd. F5 Networks
Corporate Headquarters Asia-Pacific Europe/Middle-East/Africa Japan K.K.
info@f5.com info.asia@f5.com emeainfo@f5.com f5j-info@f5.com

© 2009 F5 Networks, Inc. All rights reserved. F5, F5 Networks, the F5 logo, BIG-IP, FirePass, iControl, TMOS, and VIPRION are trademarks
or registered trademarks of F5 Networks, Inc. in the U.S. and in certain other countries. CS31316 0909

The ROI of Application Delivery Controllers in Traditional and Virtualized Environments

More Related Content

What's hot (20)

Similar to The ROI of Application Delivery Controllers in Traditional and Virtualized Environments (20)

More from FindWhitePapers (20)

Recently uploaded (20)

The ROI of Application Delivery Controllers in Traditional and Virtualized Environments