Blue Coat user Michael Shisko, the director of IT for Hitachi Consulting, said his company started looking for WAN optimization gear when it switched from an Internet VPN to a hub-and-spoke MPLS WAN service. Shisko said while 60% to 70% of the company's VPN traffic was ineligible for optimization because it was to and from Web sites, 100% of the MPLS WAN traffic was eligible. The company tested WAN optimization devices from Silver Peak, Riverbed and Blue Coat.
"The response was a remarkable dud," he says. Tests showed dramatic statistical improvement in performance, but nothing that would appease disgruntled users. Typical Word and Excel response times dropped from 30 seconds to 20 seconds, but that was still too long. "[Users] didn't perceive it as being any faster. The problem was that they weren't getting LAN-performance speed," Shisko says
Performance gains among U.S. sites weren't significant enough to warrant buying the gear, but the company installed it between its data center in Dallas and European offices cutting response times on key applications down from several minutes to several seconds.
Keeping an eye on the business goal of improving performance so end users would notice, the company bought Blue Coat optimization gear just for the international sites but not others. Hitachi Consulting already used Blue Coat's traffic monitoring/shaping gear, Packet Shaper. "The Blue Coat strategy included tight integration with PacketShaper. Just raw acceleration was not the goal but intelligent acceleration was," he says.
The goal was different for Booz & Co., which was centralizing branch servers to reduce capital and operational costs, says Ted Theofanos, senior manager of IT infrastructure for the firm.
The company tested several vendors' gear - Theofanos wouldn't say which ones - but liked Cisco's WAN optimization gear WAAS because it tunneled traffic in separate TCP sessions, enabling more detailed monitoring of traffic over the WAN. It was also integrated with Cisco's Netflow system that gathers further traffic information that can identify user and applications that are blocking more important traffic.
The gear reduced traffic between 20% and 40% of the volume it took up before optimization, and the result is that the company hasn't had to expand bandwidth on its MPLS network as fast as it would have without the technology, he says.
Similarly, Columbian Chemicals of Marietta, Ga., was doing server consolidation, says Eric Mermelstein, the enterprise infrastructure architect for Columbian Chemicals. After testing boxes from Cisco, Expand Networks and Riverbed, company decided on Riverbed. The gear is now installed at 18 sites over a meshed MPLS WAN from Verizon, allowing the elimination of 14 local servers and their associated costs.
Active directory servers still reside in branches on virtual machines in the Riverbed boxes themselves via a feature called Riverbed Services Platform (RSP) - disk space that supports VMware virtual machines. That keeps down the amount of hardware needed at each site.
Beyond meeting the server-consolidation business driver, WAN optimization boosted application performance by reducing congestion, Mermelstein says. At sites that were running at 100% capacity during business hours, WAN utilization has dropped to an average between 50% and 60%, he says. The firm is rolling out VoIP as well and needs some bandwidth headroom so call quality isn't affected by data traffic, so the reduced WAN traffic helps.
The company has also changed its disaster recovery strategy so it now replicates about 600GB of data per night between its main data center in Arkansas and the disaster-recovery data center in Georgia. Without WAN optimization, that would have taken 48 hours, he says.
Improving the disaster-recovery was the key business driver for Texon, a firm that specializes in the transportation of crude oil and tracking how much is coming in and how much is going out, says Sean Brown, senior network administrator for the firm.
The company wanted to speed up data replication between its data center in Houston and the disaster recovery in Conrow, Texas, site about 100 miles away. "They were slower than what we really wanted," he says. With a 6M dedicated link between the sites it took four days to replicate 40GB of data, he says. "We wanted more out of it, basically," he says.
Before it started using optimization software from Certeon, the company could deal adequately with the threat of hurricanes, which come with days of advanced warning. Now it can deal with all emergencies, even those that come by surprise. "We are as real-time as we can be in this situation," Brown says.
With the Certeon aCelera software in place, replication is run every day and completed in about eight hours. Some data is replicated even more often, which means there is less time for massive amounts of data to build up so the volume per day is cut in half to 20GB.
Exchange is replicated 10 to 12 times per day, just about every hour during business hours. Most of the business relies on SQL data, which is now replicated every 15 minutes or so, meaning that in a disaster, the worst case is that only the last 15 minutes of the most critical data will be missing. "To us, that's the most important aspect of our business," he says.
These WAN optimization customers have a range of advice for implementing WAN optimization systems:* When testing in your own environment, using a range of sites to get data that is representative of varying conditions.* Analyze your network traffic first to figure out what WAN performance each type of traffic requires.* Minimize encrypted traffic because it cannot be optimized as well as unencrypted traffic.* Check out vendor best-case performance claims in your environment.
Read more about lans and wans in Network World's LANs & WANs section.
This story, "Users find the secrets of WAN optimization" was originally published by Network World.