What it Takes to Be Fastest

Inverse Technology ranks America Online as the fastest ISP for loading popular Web pages. But AOL uses a cache, and that could skew the results. By Chris Oakes.

What's the fastest way to get a Web page through a modem connection? A recent measurement of Internet service providers says the trick is simple: Use America Online.

That's the message of Inverse Network Technology's latest monthly ISP survey, which tagged an AOL account as the best way for dialup users to get fast-loading Web pages.

While Inverse's test was sound, the methodology behind it may have favored AOL and similar ISP networks that use a special network feature called Web caching. Justin Newton, public policy director for the Internet Service Providers' Consortium said this qualification should be more clearly identified in the survey results.

"None of these tests are flawed as long as the test is only claiming to test what it is actually testing," Newton said. "The problem with this test is that it says it's testing end users' performance experience -- when it's only testing end users' experience for the 20 most popular sites."

By using only popular Web sites, Inverse's measurement plays strongly into the hands of ISPs that use Web caches, Newton said. Caching speeds delivery of only the most popular content, so the test "might not be indicative of overall performance," Newton said.

Web caching strategically positions popular Web pages geographically closer to end users' modem connection. Instead of retrieving a page or a graphic all the way from, say, Wired News' San Francisco Web server, a user's browser will nab it from a local cache.

The survey used 20 of the Web's most popular sites to rate the performance of 38 mostly national service providers in December 1998. The sites -- visited through dialup connections in 42 metropolitan regions -- included Yahoo, Schwab, and MTV.

Using such pages as the sole criterion leaves out delivery times for noncached pages, Newton said.

"Many of the sites [I visit], such as Hyperreal, Cloud Factory, Internet Service Providers' Consortium, ... and tens of others, will not be cached. And one wonders what the performance will be like then," he said.

Manthi Nguyen, director of benchmark products for Inverse said, "Any time you do a study, you take a sample. We are taking one slice and providing one viewpoint." By focusing on popular sites "the experience [of the measurement] would be the popular experience."

One reason for focusing on popular sites is because they're easier to define than unpopular sites. Future measurements incorporating less popular sites is a possibility, she said. "We're certainly willing to consider that if the information would be helpful to end users."

The primary goal of the survey is to provide a concrete set of numbers to the industry. In fact, details of the survey are sold to many of the ISPs that were measured.

This was the first test in which Inverse used a Web page's time in downloading as its primary metric. AOL did well on this score despite an average throughput time. Throughput time is a measurement of the number of kilobytes of data transmitted per second to an end user. AOL was only slightly better than the industry average of 2.79 Kbps.

AOL upgraded its caching abilities last September with Traffic Server, large-scale caching systems from software vendor Inktomi. The caches harboring the Web pages are deployed in AOL's data center in Virginia.

"Everyone [on AOL] goes out to the Web from Virginia," said Kevin Brown, spokesman for Inktomi. With the caches deployed, "there's a very good possibility that the object you're asking for is sitting there right in front of you."

Joe Barrett, America Online's vice president of Internet operations, said the company "knew our time to render pages was very good, but it's the first time there's been publishable results -- meaning results that we can talk about publicly."

Newton didn't dispute that AOL has clearly done the right things to improve access for its users and believes the test accurately showed that.

Nguyen pointed out that test results typically change from month to month. Variations in popular content -- whether people want streaming media or static pages -- and network backlogs can change results between surveys.

The results bode well for the state of consumer Net access, she said.

"Everybody is performing very well. The Internet service providers have really done an excellent job of moving the whole of Internet usage forward by improving the infrastructure."