When Andrew M. Odlyzko investigated why the Internet connection from his home in Minneapolis to his office just over two miles away was so sluggish, he found the answer: the data was taking a round trip to Denver.
Professor Odlyzko, a mathematician, studies the speed and quality of Internet service at the University of Minnesota. His team of researchers issues reports about the growth of Internet traffic, and he uses the team’s tools to watch his own connections, too.
Tracking the speed of Internet service is becoming more and more important as everyone asks the Internet to do more than handle e-mail messages and Web pages. A few lines of text can take its time arriving, but applications sending voice calls or streaming video become unusable if there is too much delay in delivery.
Some Web sites and software packages let users test the speed of data through their Internet service provider, or I.S.P. All the providers offer a glimpse at the quality of the connection, but that information is just one bit of data; each new request for a Web site or a file involves dozens of computers, and any of them could be a weak link.
“Even in Web browsing, pages are getting more complicated,” Professor Odlyzko said. “You click on a link and you end up setting dozens of connections. Ads are being served. You end up doing a database lookup. Any extra latency gets compounded because you have many, many stages.”
The simplest test involves downloading a big file, counting the seconds, and then dividing to compute the raw bandwidth in megabits per second. Most service providers offer a tool so their customers can test the performance inside the network, the last link in the long chain from a computer on the desk to a service provider’s distant server. Checking this last connection will usually pinpoint any problem that a service provider can fix. If the provider cannot speed up a slow connection, a user might want to make a switch.
Brett Glass, the owner of Lariat, an I.S.P. in Laramie, Wyo., said that many customers did not trust their service provider’s test and instead worked with other speed tests available on the Web. “There are a number of providers that put their thumb on the scale,” Mr. Glass said, giving a higher priority to tests for speed. Using an independent test will not isolate the last link, but it will reveal information about how well the service provider is connected to the larger Internet. Several independent Web sites can provide speed results, but the results can vary drastically.
Gary Doda works as a troubleshooter for the service provider Bright House in St. Petersburg, Fla., and owns a broadband-speed tracking site called ISPGeeks.com. Mr. Doda said many of his competitors used overloaded servers and overburdened lines linking cities and continents — called backbones — that skewed measurements. In one set of tests, he found that measurements from different sites could easily vary by as much as 50 percent to 80 percent. He suggested users try several tests and view the tests and the results with a some skepticism because Internet connections vary worldwide.
Some sites offer multiple tests bundled together. Toast.net, for instance, lets the user try downloading a variety of files from different servers scattered throughout the world that gives a better impression of how performance may vary. There are also more sophisticated tests. VisualWare offers some basic tests free online and also sells a more complex suite of testing tools at prices beginning at $30.
One measures the variation of bandwidth over time to reveal whether a service provider is slowing down or speeding up connections. Some service providers like to restrict the flow of data after a few seconds, favoring users browsing Web pages over those downloading large files.
VisualWare also simulates a voice call similar to those made on applications like Skype or Vonage. These place entirely different demands on an Internet connection because a phone call cannot survive the sporadic pauses that might not bother someone trying to download a new software package. Among other things, the test measures the amount of data lost along the way. A little lost data can be resent, but losing too much causes gaps in conversation.
On Pingtest.net, the user’s PC sends a small request to distant servers and then computes the average time to receive a response. This measure of the latency — or delay in a connection — is often a better predictor of how snappily your computer will operate, because many Web pages are assembled from multiple small files, not one big one. Pingtest offers a letter grade evaluating the quality of service.
“My dad doesn’t want to talk about latency and he doesn’t have to know,” said Mike Apcar, the chief executive of Ookla, the creator of the ping test. “So we created a grading system. If you come to our site, you get a grade A through F. Most people are going to get a B or C, but occasionally I’m having some trouble with my cable connection and I’m seeing that I’m getting a D.”
Some users are also setting up Web sites to share test results so others can see the general performance of an Internet service provider. DSLReports.com and BroadbandCensus.com, for instance, both offer tests and collect their results to produce reports to help consumers decide whether problems are sporadic or part of a larger pattern.
Unfortunately, gathering information about speeds from distant Web sites is not always helpful in diagnosing or fixing problems. The service providers have direct control over only the last link to the home, and they often cannot do much about congestion on a trans-Atlantic cable or about a sluggish server. Any difference in the results of a local test offered by the I.S.P. and an independent test could indicate whether the service provider or Internet at large is responsible for the delay.
Still, service providers have some control over how data is routed through the Internet at large, a complicated technical process called peering.
Professor Odlyzko said he was overjoyed to find out that his local service providers had set up connections with the University of Minnesota’s campus network to speed the movement of data.
“I had nothing to do with it,” he said. But he is happy that his data is no longer going to Denver.
