Accurate Broadband Speed Tests

Mike in Texas hath wroth:

Are you testing your broadband ISP connection speed, or your wireless device speed? The methods are quite different. Assuming the ISP speed, it really depends on where you are located and who is your ISP. The best test is usually a speed test server co-located at your ISP's server farm. You want the fewest number of hops between the gateway server and the test server.

Try these indexes:

(list of servers near bottom of page)

That's what many ISP's use. They also have a nifty VoIP tester. However, there's no need to have a client running on your machine reporting the speed. The application can be delivered via Java or Flash, without needing much more than a web browser. Howver, if you want your very own speed testing client, I use IPerf:

There are a few IPerf servers available on the internet, but they're primarily for the use of the ISP's customers.

You will get variations no matter how you test or what tools you use. The internet is a shared medium. In order to get repeatable results, you'll need exclusive use of the bandwidth between your machine and the speed test server. That's not going to happen once the packets leave your ISP. It *MIGHT* happen if the test server is co-located at your ISP's facilities.

I am using visual ware to check my Brighthouse/Earthlink Cable modem connection. I get good speed, but I am getting consistently lousy quality of service measurements (nothing over 65% and most approaching

30%). When I look at the route, it is always around 18-20 and the ones with the most lag and packet loss are always the same 2-3. So, is this really just measuring the QoS of the path and I might get VERY different QoS measures at one of these others?

Thanks for the great information!

Well, I was interested in ISP connection speed, but I would also like to test my wireless device speed. Don't know how to do that.

I've made some changes to my old connections in the last couple of weeks, including replacing a 25 ft. cable with a 6 ft. cable from jack to modem, replacing my 8+ year old modem with a Motorolla SB5101 SurfBoard, adding a Linksys WRT350N router and a Linksys WMP300N PCI card, and finally upgrading my service from 7Mbps to 10Mbps. My only weakness is, I'm concluding, though not directly network related, is my

500Mhz processor. But - my wireless has been running very smoothly.

I'm getting consistent numbers with each type of test I conduct, but the numbers vary with each test. Though I do get different speed results at various times of the day and night.

I have roadrunner service. Using MySpeed PC Lite, I've gotten results during different times of the day and evening between 2.8 and 9.1Mbps (4 in the morning) - with all network applications (that I know of) shut down) and Zone Alarm off. With web-based speed tests, cnet specifically, I usually get a steady 2.5 to maybe 5.0Mbps - not a great of spread as MySpeed.

I'll check out their site for bandwidth testing, as well as the other links you have suggested.

Thanks again -

Mike

Mike in Texas hath wroth:

Well, the ISP speed test requires some kind of online speed testing site. I listed a few but there are plenty others. Check with your ISP if they have one running on their servers. Most do, but usually don't advertise the URL.

Checking wireless speed is easy. It takes two computers. Unplug your cable or DSL modem. Connect a fast computer to the WAN(internet) port of your router. You may need a cross-over ethernet cable. This will be the server. You'll need IPerf:

On the server, run: iperf -s

The 2nd computer is your client. Start with an ethernet connection, not wireless. Just plug into one of the router LAN ports. Run: iperf -c ip_address_of_server You should get a speed test report in one direction. For both directions, run: iperf -r -c ip_address_of_server Lots more options to play with. See the docs.

For a sanity check, see the WAN->LAN performance charts at:

Once you have the wired benchmarking working, switch to wireless. Unplug the LAN ethernet cable from the client machine and connect via wireless. Same command: iperf -c ip_address_of_server You should get about half the wireless connection speed. If you're connected at 54Mbits/sec, you should get 25Mbits/sec thruput in both directions.

Good selection of hardware. I don't think the 25ft of cable to the modem is going to make much difference. Login to the modem at: http://192.168.1.254or: http://192.168.100.1and check the diagnostics for any RF issues.

That's not suprising when testing over a cable modem connection. As I mentioned, it's a shared medium. It's affected by what your neighbors are doing. For example, if your netmask is: 255.255.254.0 then you can have up to 1024 users sharing your cable going back to the DSLAM. It used to be possible to get better performance at night when users are unlikely to be online. However, the prevalence of file sharing, which can be run at any time, has made midnight performance less spectacular.

The 500MHz processor will not be a problem for raw downloading. It can easily handle a decent 100Mbit/sec ethernet card at wire speed. However, it may have problems dealing with some of the online applications that require substantial horsepower.

Well, if you're paying for 10Mbit/sec service, getting 9.1Mbits/sec is decent. However, you will not get that speed at any random time.

That's because the CNET bottleneck is not on your local cable connection. It's somewhere along the backhaul, which is probably constipated, throttled, or otherwise obstructed. You want the test server to be as close as possible to your ISP, prefereably on your ISP's servers.

Kurt Ullman hath wroth:

Numbers please? Your idea of good may not be the same as mine. Also, what speed service does Brighthouse/Earthlink offer?

Pardon my igorance by I think service quality and QoS (Quality of Service) are different animals. My guess(tm) is that your numbers are a conglomeration of various imparement numbers, such as packet loss, latency, jitter, and fragmentation.

18-20 whats? 2-3 whats? Packets, servers, hops, systems, etc?

Sorry, no clue. I have never seen a numerical measurement for QoS. The closest approximation are the various VoIP testers, which measure the same things, but offer individual results instead of a conglomerated metric.

It is a cable and I when I check it through Visual Ware it at least what they say cable modem should be (around 4 Mbps) and occasionally up around ethernet (10 Mbps).

Not sure what it is either (that is part of my question, I guess) those are the numbers that Visual ware gives me.

Sorry hops

Servers.

This is what Visualware gives me when I do the VoIP testers. The just say it is a measure of whether the connection is able to produce a constant stream of data. Well thanks for the time. I was hoping that these numbers were a standard and would make sense to you, I guess they don't and aren't. I appreciate the efforts.

Kurt Ullman hath wroth:

The Visualware tests explanation the various numbers.

Too many hops. You will not get an accurate speed test with that many hops. However, the VoIP latency, packet loss, and jitter test (QoS) is a good example of the voice quality you'll see running VoIP telephony. You'll find that it's quite different with a wired ethernet connection and a wireless connection. The wireless connection will have some packet loss (due to interference, reflections, and weak signals) which will increase the latency and jitter.

Well yes, there has to be a bottleneck somewhere. As I mumbled, it's usually on the backbone(s) between server farms.

Jeff Liebermann hath wroth:

I left out some details.

Both IPerf server and IPerf client should have a static IP address assigned. The WAN interface of the router should also be setup with a static IP address. Do NOT use the same subnet for both the WAN and LAN sides as this drives some routers nuts.

The first test should be with both the server and client plugged into the ethernet LAN ports and running at 100Mbit/sec (100baseT). You should get at least 70Mbits/sec in both directions. If not, something is broken. I've seen ethernet switches in some cheapo routers do weird things, so be prepared for some suprises.

The 2nd test should be with the server plugged into the WAN port. The static IP address of the server will need to change. The WAN->LAN performance tends to be insipid and much less than wire speed. It depends on the router processor speed, and the number of features, filters, and rules setup in the router.

The 3rd test is to replace the client ethernet connection with a wireless connection.

Kurt Ullman hath wroth:

Here's what my report for my home Cruzio 1500/384 DSL line looks like:

Note the packet loss at about 6.5 seconds into the test, which resulted in a huge latency value. I'm not sure what caused this hickup, but it's probably one of a half dozen assorted daemons I have running that eat CPU cycles and suck bandwidth (Skype, GizmoProject, Pidgin, NTP, etc). This hickup later caused the QoS percentage to drop well below 100%. The values much be constant to get a good QoS value. Nice test.

Thanks. now all I gotta do is figure out what all that means (g).

>

It's pretty simple, you can't get accurate speed measurements when you're dealing with multiple networking paths. Wireless adds to the complications because it can't guarantee as reliable and predictable a connection as you'd get with a wired one. Interference or other wifi traffic makes testing it difficult.

Simple web-based "speed tests" can't show you bottlenecks between your computer and the web server hosting the test.

When you're testing you want to measure the speed of your local network, the speed to the ISP and then the speed out of the ISP. The speed TO the ISP is what your service provides. But it's useless to have a fast connection to the ISP if their upstream connections out to other hosts are swamped (aka cable companies).