|
In this report we examine how the U.S. Defense Contract Management Agency (DCMA) solved application performance problems that resulted from server consolidation. As organizations centralize and consolidate application servers, IT departments must include time and budget for network upgrades that go beyond increasing bandwidth.
The DCMA is responsible for working with the defense industry to ensure that systems, supplies and services are delivered on time and at projected costs, and that they meet the performance requirements of the U.S. Department of Defense. The DCMA's headquarters is in Fairfax, Virginia, and the agency employs approximately 10,000 civilian and military professionals worldwide to manage over 300,000 prime contracts valued at more than $1.1 trillion. The DCMA works with over 16,000 contractor companies. It is organized into six divisions: Aeronautical Systems, Naval Sea Systems, Ground Systems and Munitions, Space and Missile Systems, International, and Special Programs. The divisions contain a total of 60 Contract Management Offices (CMOs), spread across more than 900 employee duty stations worldwide.
To reduce costs and management overheads, the DCMA recently consolidated its IT infrastructure in the continental U.S., from 17 data centers to two primary data centers. The consolidated applications include Microsoft Windows remote file access (Common Internet File System [CIFS]), Exchange (Messaging Application Programming Interface [MAPI]), and Secure Sockets Layer [SSL]-encrypted Web applications.
An explicit goal of the consolidation was to maintain the end-user quality of experience for all users, at local-area network [LAN]-like levels, even when applications were accessed over the DCMA's WAN. As part of the consolidation, the DCMA upgraded its WAN bandwidth in an attempt to maintain application performance at remote sites. However, immediate and very vocal end-user feedback delivered a message that slow response times were now the norm and user satisfaction was very low.
The Challenge
The DCMA had fallen prey to conventional thinking that remotely accessed application performance would be bound by network bandwidth. This was the case with client/server applications, particularly when the server was located on the same continent. However, the DCMA's applications are accessed via very chatty protocols that are highly sensitive to network latency.
Approach
The DCMA recognized that WAN optimization would have to be deployed quickly to compensate for the combination of high latency and chatty application protocols — in particular CIFS and MAPI. To simulate its WAN in a non-production environment, the DCMA used a Shunra appliance to simulate the latency and bandwidth of various T1- and T3-connected remote offices. These tests were performed for each brand of optimizer evaluated.
The DCMA compared the performance of Juniper Networks, Cisco and Riverbed devices on those simulated links. The performance of the Riverbed device stood out from the others, particularly with e-mail (MAPI) traffic. Given the importance of e-mail performance to the DCMA's customers, this was a key criterion. Riverbed also had other important key features, such as manageability, ease of installation, scalability, and the ability to "fail to pass-through," so that it would not reduce the reliability of WAN links.
Once it had determined which device was the best fit for its environment, the DCMA ran a pilot deployment between its offices in Cedar Rapids, Iowa and Carson, California. Cedar Rapids was one of the offices that most acutely felt the impact of the distance- and protocol-induced latency. Only 24 hours after the optimizers had been inserted, the DCMA received word from its Cedar Rapids customers that performance was as if the servers were "right next door" again.
Results
The DCMA has deployed Riverbed Steelhead appliances in 47 sites to date, both in the continental U.S. WAN and internationally. "If we had it to do over again, we'd install WAN optimizers before starting consolidation," said Mike Williams, CIO at the DCMA. "We ended up having to play 'catch up,' and doing it fast…Our users are much happier now. No one notices or cares anymore that we've consolidated our data centers," he added. "It's completely a non-issue now."
|
