Accounting for route congestion

[MKANET]

I may be wrong, but from what I can see...

There's no doubt that testmy.net does real-world bandwidth testing; and, that their ISP, theplanet.com, has several high quality interconnects to other popular networks. However, I've noticed that there still can be lower quality intermediary networks along certain routes that can be congested (depending on when you do the bandwidth test). So, while the general bandwidth for both the user's ISP network and testmy.net's ISP network can be very fast, there can still be a slower, congested intermediate network (specific to the route between testmy.net and the user's network) that can be the bottleneck. This can give you a lower speed rating than what the speed the user would generally see under any other circumstance.

The only way I can see testmy.net to give an accurate real-world speed test wouldbe to average out at least a few different speed tests with servers in different geographical locations.

[RTB]

That would not solve the issue, and would essentially give you a single average number that is better left as a set of numbers for more information regarding routing.

There are two speed graphs that I think are most important for an internet connection: a graph of speed tests to an internal server (within the ISPs network) to determine the average and variance during the day/week/year/millennium of your connection to your ISP; and a graph of speed tests to a server outside the ISPs network, to find out how good your ISP is at handling that. Latency tests can also be very useful.

You are right in that a single server can be hampered by a single bad network, but it does point to a problem that is not supposed to exist, and will affect many more servers which you want to interact with.

[TriRan]

That would not solve the issue, and would essentially give you a single average number that is better left as a set of numbers for more information regarding routing.

There are two speed graphs that I think are most important for an internet connection: a graph of speed tests to an internal server (within the ISPs network) to determine the average and variance during the day/week/year/millennium of your connection to your ISP; and a graph of speed tests to a server outside the ISPs network, to find out how good your ISP is at handling that. Latency tests can also be very useful.

You are right in that a single server can be hampered by a single bad network, but it does point to a problem that is not supposed to exist, and will affect many more servers which you want to interact with.

very well put RTB

[CA3LE]

Yes, very well said RTB... having said that though, I do plan on offering a test that does this. Coast to Coast speed test that will give you an idea of your average speed nationally. I've actually already been building the backend with this in mind so that when I'm ready it will be relatively easy to piece together. So look for this in the near future.

My provider isn't actually The Planet anymore, it's Softlayer. The Planet was awesome but Softlayer is far more powerful. They bought them out a couple years or so ago. DNS still resolves to the old name. After my next server upgrade the name should resolve correctly to Softlayer. Last time I upgraded The Planet gave me such an awesome deal it's been hard to beat it... for the longest time I went to their website and to make a comparable build it was like $1000/month. By the last time I got a new server I had been with The Planet for over a decade... so I was able to negotiate an awesome deal. Prices are coming down though so I'm thinking it's almost time for another upgrade. I'm getting at least 16 cores and solid state drives this next time around. It's going to be beast! queries so fast my users will be like

... Congested routes aren't something you want to compensate for in your results. That's valuable information... that's part of your performance. It often times doesn't effect your connection across the entire Internet... but if the congested route is close to home it can. If it's effecting your speed to TestMy.net's server.... it's effecting your speed to other sites as well. I purposely pick popular hubs that have high bandwidth. Let me put it this way, in the Dallas datacenter where TMN is hosted... there are 104,000+ servers. Some host single sites... some can host hundreds of sites on one server. These generally aren't small no name sites either, people with small sites don't tend to spend that kind of money on a dedicated server solution. ...... so, if your speed is effected to TestMy.net, you're speed is effected to other sites as well. The Coast to Coast test isn't going to be intended on compensating, it's more to give you a broader picture of your Internet speed.