During the last few years, the WAN acceleration market has become rife with quality products. eWEEK Labs tested two of the most recent releases, from Riverbed Technology and Blue Coat Systems, and found that both did a good job of optimizing bandwidth on a simulated WAN link.
When we last tested a product from Riverbed, in April 2004, the company had seven customers. Since that time, its customer base has expanded more than a hundredfold and its Steelhead appliances received high honors in the category of storage hardware in the sixth annual eWEEK Excellence Awards program.
Byte-caching capabilities were added to Blue Coats SG family of appliances only in May, bringing the hardware up to par with Riverbeds. Despite its relative immaturity in the WAN acceleration area, Blue Coat could quickly become a factor in the market because it is already well-known in the networking space for its caching and network security products.
Overall, Blue Coats SG800 appliances lagged behind Riverbeds Steelhead appliances in eWEEK Labs head-to-head tests, but they still provided a noticeable improvement compared with the performance of a nonaccelerated WAN.
We tested each solution using three practical scenarios to see how the devices would react to different workloads.
At the center of our test network was a Network Nightmare WAN simulator unit (www.networknightmare.com) that was set to run at T-1 speeds with 150 milliseconds of round-trip latency and 0.1 percent of packet loss.
The test units we received were overkill for the amount of bandwidth and client load we were throwing onto our simulated T-1 link, but they still illustrate the benefits WAN acceleration technology can afford to remote users and business partners.
We tested two Blue Coat SG800 units, each of which is priced at $25,950, and the Riverbed Steelhead 2020 appliance ($20,995) and a smaller Steelhead 1020 appliance ($12,495). Both Blue Coat and Riverbed offer units for less than $5,000 that will be appropriate for T-1-level bandwidth speeds, although we were unable to acquire them during our short testing window.
Good to Go
We set up the WAN acceleration units to bridge our data center and remote site switches to the ports of our WAN simulator. We ran our tests across both sides of the WAN to gauge the acceleration running in a bidirectional fashion.
Each test sequence was run three times: once to measure unaccelerated WAN performance, once to measure cold cache accelerated performance and once to measure warm cache performance.
The cold cache numbers show how well a WAN accelerator optimizes the delivery of new data. In a cold cache performance run, the WAN accelerator is seeing data for the first time and is relying on protocol optimization and compression to speed up performance.
During a warm cache run, data sent over the WAN has already been seen by the WAN acceleration product. Warm cache performance readings are considerably faster than cold cache readings because the WAN accelerators are serving up traffic to clients from their internal caches.
Our first test was a basic CIFS (Common Internet File System) file transfer, with a Windows XP client copying over a file from a remote Windows Server 2003 system. With CIFS acceleration, IT managers can either consolidate data at a central site or simply accelerate the movement of files among geographically dispersed offices and business partners.
For this test, we used a VBScript to initiate, execute and time the remote file-copy command.
Copying a 170MB test file over the WAN without WAN acceleration took 28 minutes and 33 seconds.
Using the Blue Coat solution, a cold cache file transfer was completed in 21 minutes and 46 seconds, and a warm cache transfer took just 14 seconds. Running the same tests using the Riverbed solution, a cold cache file transfer took a relatively speedy 13 minutes and 51 seconds, and a warm cache transfer took 25.7 seconds.
With MAPI (Messaging API) acceleration in place, IT managers no longer have to maintain e-mail servers at each of their remote sites. This not only facilitates storage consolidation but also allows IT managers to implement an e-mail archive at the core data center.
To test MAPI performance with the Riverbed and Blue Coat appliances, we ran a VBScript that launched Microsoft Exchange, sent a message with a 6MB attachment and timed the send transaction.
Running without WAN acceleration, this process took 126.8 seconds.
Riverbeds solution cut down the transaction time to 34.7 seconds with a cold cache and a mere 1.28 seconds with a warm cache. Blue Coats solution provided a significant improvement compared with the nonaccelerated process, but its performance trailed Riverbeds by a fair amount: With a cold cache, Blue Coat finished the transaction in 86.63 seconds; with a warm cache, it took 30.17 seconds.
In our final test scenario, we used the WAN accelerators in conjunction with Double-Take Softwares Double-Take data replication solution. We turned off Double-Takes built-in compression capabilities to allow the WAN accelerators to work at their full potential.
We set up file servers with Double-Take on both sides of our WAN and created a job that replicated across the WAN a 4GB VMware image of a Red Hat Enterprise Linux server and a file folder containing roughly 600MB of data.
Without WAN acceleration, Double-Take pushed traffic across the WAN at a pedestrian rate of 8MB per minute.
The Blue Coat appliance with a cold cache was able to speed the data delivery rate up to 61MB per minute. For the second run, with a warm cache, Blue Coat accelerated the data transfer to 308MB per minute.
Switching over to Riverbed, we saw a cold cache performance reading of 63MB per minute and a warm cache reading of 488MB per minute.
Sizing Up Expectations
As can be seen by our performance numbers, WAN accelerators can speed up traffic significantly. What cannot be ignored, however, is the sizable performance disparity between warm and cold caches.
IT managers need to do careful data analysis before setting performance expectations. Sites that are constantly sending new data over the WAN will likely see performance improvement on par with our cold cache numbers, while sites that primarily send modified versions of existing files will see performance similar to the warm cache readings.
IT managers need to be mindful of the fact that each WAN accelerator has a set amount of cache. In the case of Riverbeds appliances, caches range from 35GB in the entry-level Steelhead 100 and 200 units all the way up to 512GB in the top-of-the-line Steelhead 5010.
Purchasing an undersized WAN accelerator can lead to erratic performance because the cache will not be able to keep up with user requests.
These products can be deployed relatively easily, so we highly encourage IT managers to test these solutions with their own data before making a buying decision.
The next major leap for these products will be acceleration of SSL (Secure Sockets Layer) traffic over the WAN. Riverbed and most WAN acceleration products on the market do not have this capability right now, although Blue Coat does.
Leveraging its SSL proxy capabilities, Blue Coats SG800 can intercept and terminate SSL traffic—and then unencrypt and optimize it—before the data goes over the WAN.
Given the sensitive nature of encrypted traffic, we recommend that IT managers perform a thorough security assessment before implementing this type of SSL acceleration.
Blue Coats SG800 This WAN accelerator also has the ability to accelerate SSL traffic (www.bluecoat.com)
NetExs HyperIP A WAN accelerator that focuses on data replication (www.netex.com)
Riverbeds Steelhead These WAN acceleration appliances offer excellent performance and are a good choice for remote users and business partners (www.riverbed.com)
Senior Analyst Henry Baltazar can be reached at [email protected].
Check out eWEEK.coms for the latest news, views and analysis on servers, switches and networking protocols for the enterprise and small businesses.