What latency should I use to calculate RWIN ?

"Maximum anticipated latency" is an important estimated value, used to calculate the TCP Receive Window (RWIN) using the bandwidth*delay product (BDP).

You should use the maximum anticipated latency that you may experience at peak/congestion times. In other words, it is an estimate of the maximum ping/rtt/latency/delay.

If you use a latency number that is too small, and your latency ever goes higher, your RWIN will get filled and throttle your transfer rate.

On the other hand, if you use a latency number that is too large, your RWIN will be unnecessarily large as well. A large RWIN, combined with congested nodes and packet loss can have negative impact on your transfers as well.

Generally, as a very rough estimate, you can use the following maximum latencies:
US commercial/fiber, or residential broadband land line: 200 - 300ms
Western Europe residential broadband land line: 300 - 400ms
East Asia - residential broadband: 300 - 500ms
Wifi Internet: 400 - 600ms
Satellite connections ~700ms+
Even though the numbers above may seem large, remember those are maximum anticipated, not average latencies.

For a bit more accurate estimate, you can use the "Latency" test in the TCP Optimizer: you can try pinging a larger number of hosts, and with larger packet sizes (default is 32-byte packets that tend to travel faster)... I'd use twice the average latency from the Optimizer as your anticipated max latency.

See also:
SG BDP Calculator

User Reviews/Comments:

rate: avg:

by anonymous - 2007.09.16 16:27

300 - 400ms latency for European broadband sounds a lot, on my ADSL I normally get 30 ms withing the country, 50 ms to Google.

by Philip - 2007.09.17 07:31

Simply because you should use the maximum anticipated latency, not just an average. RWIN is a buffer, and making it somewhat larger assures that it will not limit your connection when latency spikes.

by solzfuens - 2007.12.06 14:19

I also wondered why the average for western europe is noted as worse than for the US. The backbone structure should be about the same.

A reason might be though, that more people in europe visit us websites than vice versa. And since of course latency for transatlantic connections is worse than infracontinental, the average latency for western europeans is worse. For example google, microsoft and such sites are hosted in the US but are visited by many europeans as well while there aren't that many european based sites that get huge traffic from the US.