• Does the test ever overestimate speed?
• There are speed tests that claim to be more reliable
• kilobit = 1000 bits per second
• Assuming my connection is running properly, what else may impact my speed

No. It is impossible for this Java-based speed test to overestimate speed. It can, however, underestimate due to prevailing "Internet weather" between you and the test site (which is why, if you are most interested in your last mile speed, finding a nearby test site is important), but it can never overestimate.

If you try a few different test servers, including some of the third-party ones that also use our Java test applet, then the highest speed reported is closest to your last mile speed.

Don't forget that protocol overhead will mean that you never reach the actual speed advertised by your ISP. In some cases, especially PPPoE connections, you can lose almost 15% of advertised speed in protocol overhead.

got feedback? got feedback?

Any feedback you provide is sent to the owner of this FAQ for possible incorporation and shown publically as well.

by edited by JMGullett
last modified: 2007-05-02 10:24:28

The fact is, 90% of the speed tests out there are junk. They provide results that are not reproduceable from run to run. There is at least one speed test (visualware) that offers ISPs a server-side program that claims to eliminate all speed test error. If your ISP is prepared to locate a speed test server inside its network, then it will give you the most accurate measure for last mile speed -- but is that the speed you are most interested in? Or are you more interested in testing whether your ISP is well connected to the Net and can give you full speed to many high performance download servers?

Either way, we have a large selection of Java-based speed tests located both in ISPs and elsewhere on our speed test list. If your ISP is one of those, then you will probably be able to test last mile speed the best by using their local test server. If you are interested in real world test results, then pick another well-connected server. It is your choice.

got feedback? got feedback?

Any feedback you provide is sent to the owner of this FAQ for possible incorporation and shown publically as well.

by edited by JMGullett
last modified: 2007-05-02 10:28:24

For the purposes of presenting speed test results, we adopt the data communications convention of k = 1000, not k = 1024. For example, 28.8k modems ran at 28800 bits per second, and 56k modems ran at 56000 bits per second.

The transfer rate expressed as kilobytes per second is based on 1024 as per data storage conventions.

got feedback? got feedback?

Any feedback you provide is sent to the owner of this FAQ for possible incorporation and shown publically as well.

by edited by Mike
last modified: 2012-02-24 09:35:40

Outside of any end to end congestion issues, there are a number of things which may negatively impact speedtest results:

1. Physical distance. You have likely noticed that downloading from a local server location is faster than a distant one. This is because the TCP window (e.g. HTTP transfers) is not optimized for the increased latency that comes from an increase in physical distance the bits have to travel. HTTP is bound by the TCP congestion window which determines how many packets can be sent at one time before an acknowledgment. The larger the window size, the higher the TCP throughput.

2. Firewalls, wireless, Local LAN, PC and modem issues. There are many localized issues which may negatively impact speedtest results. The best (but not perfect) PC test environment is to have a high end PC capable of delivering the provisioned speeds, directly attached to the modem.

3. Capacity issues with intermediate ISPs between the server and the user. There are some speedtest servers on ISPs with poor interconnect relationships or capacity issues. The path between the user and the server factors into any results. Local, on-net, speedtest servers are best to measure your provisioned speed.

4. Tuning of the Speedtest server with parallel TCP streams. This is not well known, but does have a direct impact to higher broadband speeds. Most speedtest servers open multiple TCP streams to more accurately account for single stream TCP limitations. How many simultaneous TCP streams a server opens is important as with 4 streams a server is accurate to up to 50Mbps testing on an average PC. Some speedtest servers are configured for even fewer streams and therefor are less accurate at higher speeds.

5. Asymmetric test environments - High end commercial connections with symmetric speeds often ask why they get asymmetric results from most speedtest servers. This is because most speedtest software / servers are configured for residential broadband testing. They are normally configured for 2-4 simultaneous download streams / test, but only 1-2 upload streams. This will inaccurately give results of asymmetric bandwidth even when you have symmetric speeds. Those with higher end work connections can simulate this.

got feedback? got feedback?

Any feedback you provide is sent to the owner of this FAQ for possible incorporation and shown publically as well.

by ccneteng edited by Mike
last modified: 2010-11-09 19:29:33