I recently tested a beta version of WebAvalanche in our Green Bay, Wis., Real-World Labs® and found it can pump more than 500 Mbps of traffic and sustain more than 1 million TCP connections without dozens of PCs and software agents. Many of the new features were added in intermediate upgrades, but version 4.0 pulls them all together and provides a single-unit solution for massive load generation and testing.

WebAvalanche's Web-based interface hasn't changed, but there are plenty of new options and tweaks. Now tests can be based on connections per second, transactions per second or "CawUsers" per second. A CawUser tests devices based on sessions and is designed to emulate a user as he or she clicks through a Web site.

Streaming-media testing requires little configuration and can generate thousands of concurrent streams to put the pressure on your media server. WebAvalanche provides real-time statistics for streaming tests, including average packet loss, number of streams and bandwidth distribution, response time, and concurrent active streams and channels per second.

Vendor Information WebAvalanche 4.0; starts at $19,500, options with streaming start at $39,500. Caw Networks, (408) 327-5200 (main), (408) 327-5300; fax (408) 327-5201. www.caw.com

Caw's solution has an advantage over its competitors, many of which also support streaming media, in that it requires only a single device to do what previously required many machines loaded with agents. A single 4U device with either 10-/100-Mbps or 1-Gbps fiber connectivity provides for lower maintenance, installation time and cost. The pricing of WebAvalanche is steep, but compare it with the licensing fees of a software-based solution -- and don't forget to include the cost of the hardware and maintenance. WebAvalanche comes out on top in that equation.

Screaming...

Upgrading to the newest version of WebAvalanche requires popping out the old flash card, inserting the new card and rebooting. WebAvalanche, like its server-side companion, WebReflector (see a previous Sneak Preview on CAW's WebReflector), uses QNX as a bootstrap loader only, employing it to load Caw's proprietary OS from the flash card.

The biggest complaint I have about upgrading is that all my test configurations needed to be updated. None of them worked after the upgrade because of the significant change and enhancements since version 1.0 of the device. If you have a lot of different test scripts, as we do, the process of upgrading scripts is quite a chore. I'd love an automated method for upgrading scripts included in the product.

As I was adjusting my test scripts, I noticed (and was pleased that) the time to run a test is calculated automatically based on the test parameters. I worked on a script to generate 1 million concurrent TCP connections with an initial ramp-up for 1,000 sessions and increasing every 10 seconds with 10,000 sessions. As I entered the data, the product automatically calculated that the entire test run would take 20 minutes to complete. This is a nice change from the initial version, which required all parameters to be hand entered, because it was easy to miscalculate the time required to complete a full test.

After fixing the script, I fired up the WebAvalanche and directed it to connect to two virtual Web servers being served by the WebReflector. The test ran excellently, soon reaching 900,000 sustained TCP connections with more than 6,000 new connections being created every 10 seconds. The real-time statistics window showed more than 500 Mbps of traffic flowing between the two devices, with very few unsuccessful connections occurring. This type of test is useful for benchmarking load-balancers, content switches and proxies. It also showed WebAvalanche's ability to pump out traffic.

WebAvalanche lets you designate both the MSS (maximum segment size) and the number of retransmits for TCP time-outs before the transaction is abandoned. The MSS defaults to 1,460 bytes, and when you add a 20-byte header for each IP and TCP, you get the accepted standard MTU of 1,500 bytes. Be wary of playing with this field, since setting it higher can push the MTU above 1,500 bytes, causing packet fragmentation and wreaking havoc with your network.

Caw's WebAvalanche also offers a wide variety of parameters to better simulate real users, such as user think time (time between requests), click-away time (abandonment) and bandwidth limitations as well as custom headers (preconfigured Netscape and Microsoft Internet Explorer headers are also available) and the ability to authenticate to a proxy cache.

As should be expected from a load-generation tool, WebAvalanche lets you configure a test with several different user profiles. Each profile is configured independently with its own parameters and can be used in any test script.

... and Streaming

I set up two separate Linux servers running Apple's open-source streaming server to test the streaming load-generation capabilities of the WebAvalanche. I configured a new script with two user profiles. Each profile accessed a 100-Kbps stream served from one of the two servers. I set a low threshold for the number of transactions per second to be performed -- 100 being the maximum -- and started running the script. The real-time performance tab within the statistics window offered an excellent view of the servers' performance, from number of streams open and active to packet loss to response time.

In the beta, I had to configure the script simply by pointing each user profile at a URL using the RTSP protocol. That meant that instead of using HTTP://10.1.1.90/somefile.html, I simply entered rtsp://10.1.1. 90/somemovie.mov.100k in the URL list. Caw Networks indicates that the GUI is being enhanced to include easier configuration for testing streaming servers and will be available with the final release of version 4.0.

So before you deploy that streaming video of Pamela Ander -- uh, your CIO's remarks at the annual company meeting, beating on the servers with WebAvalanche will help you ensure that you have the hardware to serve everyone without any degradation in quality.

Technology editor Lori MacVittie has been a software developer and a network administrator. Most recently, she was a member of the technical architecture team for a global transportation and logistics organization. Send your comments on this article to her at lmacvittie@nwc.com.