Study cites 'Net backbone for slowing Web sites

The study found that 35% of Web site performance bottlenecks occur outside the firewall, caused primarily by tuning and configuration issues with ISPs' routers, gateways and switches as well as peering point problems between ISPs. Another common problem is that Web sites do not have enough Internet bandwidth, either because they underestimated how much they need or because of a faulty service-level agreement with their ISPs.

"Everybody thinks that the Internet is an information superhighway, but it's really a dirt road full of potholes," says David Gehringer, senior product marketing manager at Mercury Interactive who oversaw the study.

The 35% figure is much higher than Mercury Interactive analysts guessed prior to completing the study. "We figured it would be about 5% to 10%," Gehringer says. "We figured it would be a pretty rare instance."

Mercury Interactive based its study on nearly 2,000 Web site load tests conducted for more than 600 corporate customers. Mercury Interactive published a similar report six months ago, but the company did not measure the number of performance problems found on the Internet vs. internal corporate networks and systems.

The new study found that in 25% of all cases, the Web site's connection to the Internet couldn't sufficiently handle the desired load, which caused delays for incoming and outbound requests. Web sites often pay for bandwidth that they aren't receiving from their ISPs, particularly when they have contracts that specify burstable rates.

Other common network problems that slow Web site performance are incompatibilities related to firewalls, overzealous security settings and improperly installed load-balancing equipment.

"As we're running our tests, our customers often realize that the network architecture at a very basic level is misconfigured," Gehringer says. "By changing some simple settings, we're often able to double or triple the speed or capacity of the site."

Other common Web site performance problems include:

??Database tuning problems, which caused 27% of Web site performance problems. Often found in dynamic Web sites that rely heavily on database operations, these problems are usually caused by insufficient indexing, fragmented databases, out-of-date statistics or faulty application design.

??Application server configuration issues, which caused 23% of system bottlenecks. These include poor cache or session management.

??Web server bottlenecks, which were found 20% of the time. Improper server configurations and insufficient memory usually cause these bottlenecks.

Overall, Mercury Interactive found that 98% of the Web sites it tests have significant performance problems, a figure that is consistent with the company's findings six months ago.

Typically, a site is able to scale up to only 15% of the anticipated performance level when it moves from development to production, Gehringer says. "Most of the sites have purchased the right amount of hardware to do what they want to do, but they've missed some basic settings or tunings," Gehringer says. "We're finding performance problems in some of the best sites on the Internet, ssites that are not short money or talent."

Mercury Interactive says its customers are able to improve overall Web site performance by up to 400% after conducting a series of performance tests and resolving the configuration problems that are found. When the performance problems are found outside the firewall, companies can take the data from their performance tests to their ISPs to help pinpoint and fix problems.

Mercury Interactive's findings ring true to Theresa Lanowitz, research director for applications development at Gartner.

"The application development side of the IT organization and the operations side of the IT organization usually work in silos," Lanowitz says. "The application development group does testing with respect to how that application performs standing on its own. They don't necessarily look at the larger picture of how it will perform with a large load running on the production environment."

Lanowitz says hosted testing services from such companies as Mercury Interactive, Compuware and Rational Software are useful in testing how a Web application will perform in the real world before it goes live. "When you have the hosted service set up and have people who as their professional career test applications, they understand performance, and they understand load balancing," she says. "They can look at the application and the platform and tell you where things should be because of the repetition of what.???????????.????????????? ..????????????? ..?????????????