Redline's Answer to Slow Connections

The T|X 2650 is built on a custom OS and is designed to alleviate the problems associated with servicing low-bandwidth clients. For example, Apache Web servers deployed on low-end hardware--like the PIII 600-MHz machines in our Real-World Labs® in Green Bay, Wis.--can maintain only a limited number of connections. If most of your visitors are connected via a 56-Kbps or slower modem, the SendQs available to Apache fill up quickly but empty out slowly, limiting the number of visitors the server can accommodate.

I tested the T|X 2650 in our labs, using Spirent WebAvalanche to emulate clients accessing a single Apache Web server at 33.6 Kbps. Without the T|X, the server attended to 55 clients per second, each requesting a single HTML file with three embedded images for a total of 24 KB. With the T|X, the Web server successfully served 330 clients, and the total time to retrieve the entire page dropped from 17,501 ms to 9,163 ms.

High-Wire Act

The T|X 2650 offers some basic Layer 4 load-balancing capabilities. Up to 128 clusters (pools) can be configured with up to 8,192 servers. The GUI allows configuration of up to 32 servers per cluster; additional servers in a cluster must be configured on the CLI. The device supports persistence via cookies but does not offer other configurable choices for the load-balancing functionality.

The T|X 2650 has a "busy redirection" option that redirects a browser to a specific URL when the device is too busy to process a request. The URL could try to redirect back to the T|X after a specified amount of time or show special content while the user is waiting for access to the site.Even if you have only one back-end server, you'll still configure a cluster. This lets you quickly add additional servers to a cluster if you need more back-end processing power. I added a second server to the cluster via the GUI in less than a minute and then ran tests with the WebAvalanche against the cluster. Capacity increased to 816 client transactions per second with an average page load time of 2,944 ms.

The T|X 2650 lets an admin enable SSL on the listen (client facing) or internal (back end). The cluster's port is not changed, however, so the test run over SSL ran over Port 80 instead of over the expected Port 443. Response time was a mere 1,937 ms with 885 page loads per second being served. The T|X takes advantage of SSL acceleration internally to provide better SSL negotiation and bulk encryption rates.

At What Cost Speed?

The biggest drawback to the T|X 2650 is its price. At $49,995 for a single device and optional support of $3,995 annually, it is not a midmarket solution. The increase in performance of a single, low-end machine and subsequent increases through load-balancing of multiple low-end machines is likely worth the savings associated with repurposing legacy hardware. Still, I'd like to see additional load-balancing options included to justify the price, especially when you compare the T|X 2650 to competitors whose only difference is a lack of compression technology.

Lori MacVittie is a Network Computing technology editor working in our Green Bay, Wis., labs. Write to her at [email protected].Post a comment or question on this story.