How to measure your true Internet speed

Woody Leonhard

Curious about your Internet speed? Most Internet service providers (ISPs) offer some sort of throughput test tool on their sites —just click, and you’ll get a couple of often-impressive numbers.

It’s in an ISP’s interest to provide the best speed numbers possible; your actual throughput is probably something quite different.

Understanding the nuances of Internet speed

Let’s start with a few basics. In most cases, consumers measure Internet speed three ways: download speed, upload speed, and/or latency/ping time. Download and upload speeds are usually reported in bits per second — typically, kilobits per second (Kbps) for slower connections and megabits per second (Mbps) for faster connections. (Extremely fast connections might also be measured in gigabits or terabits. Convert-me.com has a handy online tool for converting transfer units.)

Almost all Internet speed–testing sites measure how long it takes to download one file over one connection from a webserver to your PC. Depending on where and how you download most files, that might be an accurate reflection of your average throughput.

However, all major browsers have download accelerators built in. (Free, third-party accelerators are also available as browser add-ons.) Accelerators typically download files over two or more connections — often without any indication to browser users. That makes it impossible to predict whether an accelerator is delivering a faster download — and how much faster that delivery might be.

Torrents (used for downloading large files; more info) take multiple data streams as an article of faith. So your typical Internet speed test probably won’t provide an accurate measure of torrent download speeds, which depend on mostly uncontrollable factors such as the number of sites (or seeds) offering the file for download and the speeds of those seeding sites. (This assumes your ISP doesn’t block torrents — more about that later.)

As with download-speed tests, upload tests measure how fast it takes to send one file over one connection from your PC to the cloud. Depending on your Internet service, upload speed can be one-third to one-tenth your download speed. (For example, the “A” in an “ADSL” connection means “asymmetric” — which, in almost all cases, is optimized for downloads at the expense of upload speeds. Some services — typically for business — provide similar upload and download speed.)

Ping or latency tests measure how long it takes for one small packet of data to travel from your computer to the testing site and back to the PC. Ping times are almost always measured in milliseconds (ms; thousandths of a second).

In short: Download speeds are important when downloading lots of small files or streaming large files such as video. Upload speed is important when sending large files to the cloud (such as uploading big photos and video) or when using Skype, live video cams, and video calls. Latency comes into, uh, play with online, interactive gaming — where a delay could leave you fatally vulnerable to zombie attacks.

When the Internet was relatively new, measuring connection speed was a sort of propeller-head pastime, run when we had a few spare minutes — kind of like reformatting the C: drive, only much faster. Today, the Internet is essential to our daily computing and to much of our entertainment. As we’ve become more sophisticated consumers, we want to know that we’re getting most of the connection speed our ISPs advertise. I’d be really disappointed if I bought a car advertised to reach 110 mph, only to discover that I needed a steep downhill and a significant tailwind to do so.

Some ISPs genuinely want to provide connection speeds their customers expect — hard to believe, I know — and their online, speed-test tools help them determine the rates they’re actually delivering. Many ISPs provide something far less than promised, resulting in lawsuits all over the world.

Tests that report intentionally inflated results

When you use an ISP’s speed test, you’re testing only the relatively short connection between your computer and the ISP’s server. And even at that, it’s probably not anything like a real-world connection; some ISPs skew the results by sidestepping their usual Internet routing. The result is a foregone conclusion: lightning-fast scores. These bogus tests have possibly helped bring down the legitimate wrath of ISP customers, and I hope — hope! — this kind of fiddling is on the way out.

There are other ways that ISP customers might get a somewhat skewed sense of their Internet connection speed. For example, almost all ISPs serve up popular webpages more quickly by caching pages on their local servers. If, say, a 100,000 Internet users in Seattle are watching their football team play in Chicago, the ISP doesn’t have to move 100,000 identical sets of digital video bits from Chicago to Seattle.

Caching’s great for many reasons, but if you’re trying to measure your typical Internet speed, it can gum up the results — especially if you’re using third-party, connection-speed test sites. When a latency/ping test comes in faster than the speed of light, either there’s something fishy going on or your ISP has learned how to exploit a wormhole.

A perfect Internet connection should travel around the world in about 13.3 milliseconds (the speed of light). Real connections run much slower because they’re routed through numerous servers and switches. If you run a speed test on a site halfway around the world and the latency comes in under 25 milliseconds or so, you’re undoubtedly looking at a cached test.

Another wrinkle I’ve seen with third-party test sites: Some ISPs, noting which outside test sites their customers are pinging, produce better test results by optimizing the connection with those sites. If you run the same test on a different site and get much slower results, you’ll know why.

Different testing sites, vastly different scores

Ready for an eye-opener? Try this quick experiment. Run speed tests on all the following, well-known sites and see whether there’s any consistency with your scores.

The most frequently used test is on Ookla’s Speedtest.net site. You must have Adobe Flash installed to run the test, and the site pushes you to download sponsoring apps, such as Google Chrome. You can pick available servers, world-wide, with the site’s on-screen map.

Using Speedtest.net, I connected my Thailand-based PC to a server in Los Angeles, Calif. I clocked a download speed of 9.0 Mbps — not bad for an ADSL line rated at 10 Mbps.

Another site, DSLReports, shown in Figure 1, is also Flash-based. I’ve used it for years. My download score on this site was 1.5 Mbps. That’s one-sixth as fast as Speedtest.net — same computer, same line, but very different results. There’s a problem, eh? DSLReports also has a Java-based test, which produced a nearly identical score of 1.4 Mbps.

DSLReports

Figure 1. DSLReports running its Internet speed tests.

MLab’s Network Diagnostic Tool is an open-source, distributed testing service that runs on servers all over the world. Sponsored by BitTorrent, Google, Amazon Web Services, Microsoft-owned Skype, and many other commercial and academic institutions, it’s an attempt to provide unbiased, research-oriented, test results. The site does not let you choose a specific server. My score: 7.3 Mbps.

In the oldie-but-goodie category, CNET’s Flash-based Bandwidth Meter uses a predetermined, but unidentified, server somewhere in the U.S. and a small, 1.5MB, download file. My score: 8.7 Mbps.

For a peek at an HTML 5–based site, check out the Bandwidthplace speed test. This site lets you easily choose servers in the U.S., Europe, and Asia. (I chose a server in San Jose, Calif.) My score: 1.1 Mbps.

Visualware has a compelling test site. It isn’t as pretty as the others, but it doesn’t require Flash and the results chart is a bit more informative (see Figure 2). My score (connected to a Los Angeles server): 1.9 Mbps.

Visualware

Figure 2. Visualware's Internet test site includes an informative performance chart.

So what’s my true connection speed? Does my line run at 9.0 Mbps (Ookla), 7.3 (MLab NDT), 1.9 (Visualware), 1.5 (DSLReports), or 1.1 (Bandwidthplace)?

I took all those measurements on my main production machine — a basic system with Windows 8 and IE 10 installed. Could the browser or the OS have contributed to the squirrelly results?

To check, I reran the same tests on a Win7 PC (all patches applied) with Chrome Version 23. The results were all within 15 percent of the Win8/IE 10 scores — some a bit higher, some lower.

Clearly, the discrepancies lie with the testing sites.

No one site provides a clear answer

Given the wide range of results from the various test sites, which test should you use? I’m sorry to say, there’s no definitive answer.

The U.S. Federal Communications Commission, which should be able to make the call if anyone can, waffles. It offers the Java-based, Consumer Broadband Test (Beta) on its Broadband.gov site. Just the fact that you must enter your full address is enough to put anyone off the test. And the FCC doesn’t even try to adjudicate which of the existing tests is the most accurate. Here’s how it punts on the topic:

  • “Because measuring broadband speeds with software tools is not an exact science, we are providing two popular consumer broadband testing tools in this Beta version: Ookla and M-Lab. Both will enable consumers to test the quality of their broadband connection by transferring a small temporary file back and forth and measuring the results.

    “Users will be randomly assigned to one of the two chosen testing tools: Ookla or Network Diagnostic Tool (NDT) running on the Measurement Lab (M-Lab) platform, or they can choose their preferred tool by using links on this page. Each test is likely to provide a different result, and the differences may be significant in some cases.

    While the tests will give consumers some information on relative speeds, the FCC does not endorse either one as being a definitive testing method. In the future, the FCC anticipates making additional broadband testing applications available for consumer use. The FCC does not endorse any specific testing application.”

That leaves us up the ol’ creek without an officially endorsed paddle. But I do have two observations.

First, the tests I used are all surprisingly replicable. If you run the Ookla test on two different machines, attached to the same line, you’ll get roughly the same result. If you run the DSLReports Java test now, and then run it again 10 minutes from now, again you’ll get roughly the same result.

So the major benefit of running speed tests isn’t to get an absolute speed number. Instead, you should be looking for variations. Does your line slow down every weekday at 4:00 p.m., when kids come home from school, or at 9:00 a.m. weekdays, when the business day starts? Does it slow down on weekends or speed up in the middle of the night? Does your line fall on its knees when it rains? These tests will give you reliable answers to those kinds of questions.

Second, none of the tests listed above will definitively tell you whether an ISP is giving you significantly less speed than you signed up for. (You might get a clue if all the testing sites reported scores well below what you’re paying for.)

There is, however, a different kind of test that’s potentially worthwhile — and it’s more than a little disconcerting. MIT’s Max Planck Institute for Software Systems has a multifaceted test — Glasnost — that’s designed to “make ISPs’ traffic-shaping policies transparent to their customers.” The tests measure and compare data traffic that’s commonly throttled by ISPs: torrents, e-mail, common HTTP Web access, newsgroups, and Flash-based content (e.g., YouTube). You can run the tests directly from the Glasnost site. Each test takes about eight minutes.

The whole concept of “net neutrality” — preventing ISPs from playing favorites with certain kinds of data or giving preferred access to certain sets of sites — opens an entirely different can of Internet-speed worms.

Fortunately, in the long term, Internet speeds should not be as crucial as they are today — especially if Google can be convinced to extend its U.S. $75-per-month, 1,000 Mbps (up and down) Google Fiber service (site) beyond Kansas City. But even if speeds improve dramatically across the board, the thorny questions of net neutrality remain.



Subscribe to our Windows Secrets Newsletter - It's Free!

Get our unique weekly Newsletter with tips and techniques, how to's and critical updates on Windows 7, Windows 8, Windows XP, Firefox, Internet Explorer, Google, etc. Join our 480,000 subscribers!

Windows 8 Hacks: Tips & Tools for Unlocking the Power of Tablets and Desktops

Subscribe and get our monthly bonuses - free!

Want to hack the new Start screen and tiles for your Win8 Device, the new Lock screen, the new tile-based apps, or the automatic notification information? Yes, you can do that. How about running other operating systems inside Windows 8, running Windows 8 on a Mac, or hacking SkyDrive and social media? We'll show you how to do that as well. Get this excerpt and other 5 bonuses if you subscribe now!

= Paid content

All Windows Secrets articles posted on 2012-12-06:

Woody Leonhard

About Woody Leonhard

Woody Leonhard is a Windows Secrets senior editor and a senior contributing editor at InfoWorld. His latest book, the comprehensive 1,080-page Windows 8 All-In-One For Dummies, delves into all the Win8 nooks and crannies. His many writings tell it like it is — whether Microsoft likes it or not.