Networking: The Need for WAN Optimization
By Jim Metzler
September 15, 2008 — A lot has been written about WAN optimization over the past three years, as companies have discovered significant performance problems when branch office workers try to access applications that are housed in centralized data centers. A host of factors has come into play to complicate the task of ensuring acceptable application performance in the branch. Two of the primary organizational factors are the movement to consolidate IT resources into a smaller number of locations and the simultaneous movement to distribute employees geographically.
Many companies either already have consolidated or are in the process of consolidating servers out of branch offices and into centralized data centers. This consolidation typically reduces cost and enables IT organizations to have better control over the company’s data. Because of server consolidation, chatty protocols such as CIFS (Common Internet File System), Exchange and NFS (Network File System), which were designed to run over the LAN, are now running over the WAN.
A chatty protocol is one that that requires hundreds of round trips to complete a single transaction. To understand the challenges this presents, assume that it takes 300 round trips to complete a transaction, that the LAN round trip delay time is 1ms and that the WAN round trip time is 100ms. The transaction takes a fraction of a second (roughly 300ms) to complete when accessing a server on the LAN. This is such a small increment of time that no user would notice it. The same transaction, however, requires a minimum of 30 seconds to complete when accessing a server over the WAN. This is so large that everybody notices it.
To put chatty protocols in context, CIFS works by decomposing all files into smaller blocks before transmitting them. Assume a client is attempting to open up a 20MB file on a remote server. CIFS would decompose that file into hundreds or possibly thousands of small data blocks. The server would send each of those data blocks to the client, where it would be verified, with an acknowledgement sent back to the server. The server would have to wait for the acknowledgement before sending the next data block. As a result, it could take tens of seconds before the user would be able to open the file.
In addition to consolidating servers out of branch offices and into centralized data centers, many companies are reducing the number of data centers they support worldwide. Hewlett-Packard, for example, recently announced it would reduce the number of data centers it supports from 85 down to six. Many companies are also adopting a single-hosting model, whereby users from all over the globe transit the WAN to access an application that the company hosts in just one of its data centers.
At the same time that companies are centralizing their IT resources, they are distributing their key business functions (manufacturing, engineering, customer service, R&D)—and hence their employees—around the country and, in some instances, around the world.
The distribution of employees is driven in part by the need to be close to customers. Many companies have found that having employees in branch offices, instead of in a headquarters facility, increases the number of touch points with existing and prospective clients. For logistical reasons, companies might deploy both distribution and customer service centers close to their customers. Another driver is cost reduction: Companies are moving such functions as manufacturing and call centers to lower-cost areas of the country or the world.
Yet another factor driving the distribution of employees is the desire to access scarce talent. This causes companies to set up R&D facilities close to research universities, no matter where those universities are located.
The concomitant moves to deploy employees in multiple geographical locations while consolidating servers out of branch offices and into centralized data centers mean that most employees now access applications over a relatively low-speed, high-latency WAN instead of over a high-speed, low-latency LAN. That fact, combined with the movement to a single-hosting model, means that the WAN that the remote employees must use to access applications is getting longer. And typically, the longer the WAN, the greater the delay and packet loss.
A further factor complicating the task of ensuring acceptable application performance by branch office employees is the movement to implement new application architectures such as SaaS. In that model, customers do not pay for owning the software itself, but rather for using it. They use it through an API accessible over the Web and often written using Web services.
There are many challenges associated with SaaS. Since, by definition, SaaS is Internet based, the user is exposed to all of the Internet’s foibles. And since the company that uses the software does not own it, the user cannot alter the software to make it perform better.
There are many other existing and emerging factors that bedevil efforts to ensure acceptable application performance. Thankfully, over the last few years we have also begun to see the deployment of solutions that help applications run better over the WAN.
Jim Metzler has worked in just about every aspect of the networking industry in more than 30 years of professional experience. He can be reached at jim@ashtonmetzler.com.
Related Search Term(s): networking, SOA & SaaS
Share this link: http://www.sysmannews.com/link/32803