Executive Summary
When planning the capacity
of a WAN optimization solution, it’s best to base calculations on
a WAN optimization appliance’s ability to perform
application-layer functions, such as managing user counts and
protocol optimization throughput, rather than on low-level metrics
such as TCP counts. WAN optimization appliances function as WAN
proxies, changing network traffic to accelerate it and secure it.
These changes can affect TCP counts in unpredictable ways, making
TCP counts a poor choice of a metric for capacity planning. A
better approach is to determine how many application sessions a WAN
needs to support and then to design the WAN solution accordingly.
An effective WAN acceleration solution will make optimal use of
lower-level TCP functions while also scaling its higher-level
application and session-based functions to meet the growing demands
of users.
Application Layer vs. TCP Layer WAN Optimization
Choosing the Right Metrics
Consider this
scenario. A financial services firm has grown rapidly over the past
decade, and now it’s time to upgrade the network. The goal is
to
increase the network’s capacity to accommodate new employees and
branch offices. The upgrade will include a new WAN optimization
solution with WAN concentrators deployed at core data centers and
at branch offices.
To plan the capacity of the WAN solution, the company’s Network
Operations Center (NOC) engineers determine how many TCP
connections their
employees are using in their daily work. To give themselves a
margin of error, the engineers increase this number by 20%, then
multiply it by the projected number of employees the company will
have in 3 years. Then they buy and deploy their new WAN
optimization appliances.
But there’s trouble right away. Network performance is sluggish
everywhere. Even though it was designed to accommodate a much
larger organization, the WAN optimization solution is already maxed
out and not delivering the promised performance; in fact, it’s
slowing down applications.
What went wrong?
The NOC team made a
critical error in scoping their WAN optimization solution. While
it’s perfectly reasonable to use TCP connection counts as a
capacity planning metric for routers and other internal network
devices, it’s inappropriate for scoping the capacity and design
of WAN optimization
appliances. Why? Because WAN optimization appliances are
essentially proxies, intercepting LAN application traffic and
changing it – for the better –
on the WAN. And the effect of proxy services on TCP count is
unpredictable.
Improving applications can increase, or decrease, the TCP
connection load in ways that defy simple back-of-the-envelope
calculations.
A Closer Look at WAN Optimization
Effective
WAN optimization solutions manipulate traffic in a variety of ways
in order to deliver applications and data quickly and securely to
users
throughout the enterprise. Authentication, byte-caching,
compression, protocol optimization, policy enforcement, and other
proxy activities all have the potential to change the number of TCP
connections active on a WAN. Exactly how these activities change
TCP counts varies, and it varies so much that TCP counts turn out
to be a poor metric for planning WAN optimization capacity.
For example, consider the effect of protocol optimization on TCP
connections. A powerful technology, protocol optimization aligns
high-level protocol behavior with low-level network realities.
However, the “optimal” way to transmit data could involve more
or fewer TCP connections, depending on the protocol and the
circumstances. For example, a user Web request going across an
optimized WAN might have two to five TCP connections on the LAN
side, one connection on the WAN between appliances, then balloon to
50 or more connections on the far end, as the appliance leverages
parallelization to improve performance. But protocol optimization
can also do the reverse, as when a proxy aggregates CIFS file
service requests, thereby reducing overall TCP count.
TCP counts are a misleading metric for capacity planning. They’re
distracting, too. Protocol optimization, authentication, and other
proxy-like activities tax any WAN appliance far more than holding
open TCP connections do. How well a WAN optimization appliance
performs these acceleration and security functions for users will
ultimately determine the capacity of the WAN solution. By
overlooking these higher level functions and instead focusing on
TCP counts, NOC engineers risk deploying an appliance without the
horsepower to do its job well, however many TCP connections
it
can hold open.
The Best Practice for WAN Optimization Capacity
Planning
When capacity planning a WAN optimization
solution, it’s best to base calculations and scoping requirements
on user sessions and application throughput. WAN optimization
solutions need to optimize the requisite number of user sessions,
regardless of how many TCP connections result
through parallelization and other optimization techniques. Whether
a single session, such as a user running a Web application, results
in 50 TCP connections or only 5, the appliance will still have
plenty of capacity for managing TCP connections, as long as it has
enough horsepower to perform
these other higher-level optimizations.
The goal, ultimately, is to deliver low-latency, network efficient
and policy-compliant, applications to users everywhere. To avoid
network bottlenecks
and other unpleasant surprises, focus on user sessions and upper
layer metrics, rather than lower-layer metrics that ultimately
serve only to implement the higher-layer, optimized WAN
solution.
Running a pilot project and monitoring how well a WAN optimization
appliance manages its higher-layer functions is a good way to
ensure the broad WAN optimization solution will have adequate
capacity when it is deployed.
The Blue Coat Solution for WAN Optimization
Blue Coat is a leading provider of WAN optimization solutions that
accelerate business applications across the distributed enterprise.
Blue Coat’s family of appliances and client-based solutions –
deployed in branch offices, Internet gateways, end points, and data
centers – provide intelligent points of policy-based control
enabling IT organizations to optimize security and accelerate
performance for all users and applications.
Blue Coat appliances use a proxy/cache architecture that is user-
and application-aware. By analyzing higher-level application
functions, such as
authenticated user sessions, and making use of caching and session-
and application-layer techniques, Blue Coat ProxySG appliances can
optimize
application delivery far more effectively than WAN solutions whose
designs treat WAN optimization as a networking problem confined to
the packet
delivery layers of the OSI model (layers 2-4). As a result, in
real-world deployments, Blue Coat WAN optimization appliances
consistently deliver
faster performance and greater scalability than other optimization
products.
Conclusion
When comparing WAN optimization
architectures, it’s important not to get sidetracked counting
packets and TCP connections. Focus instead on
delivering applications quickly. That means looking where users,
applications and their data live: at layer 7. Measure WAN solutions
by their ability to
deliver accelerated applications, not packets, and you’ll achieve
your WAN optimization goals.
