Question about Timestamps

wangnan · Post by **wangnan** » Sun Dec 28, 2003 10:49 am

Microsoft says that "Timestamps help TCP measure round trip time (RTT) accurately in order to adjust retransmission timeouts" (http://www.microsoft.com/windows2000/te ... /58813.asp)

However I have not see anyone here use Timestamps feature for optimizing TCP/IP performance.

What I understand from MS's description is that Timestamps feature can find correct retransmission timeouts automatically, in order to reduce retransmissions. So this is good for data transmited over a very long route (For example, from Miami to Hongkong), since its transimission time varies largely and is diffcult to predict. However, in a LAN, we should turn off Timestamps feature.
Am I right?

My question is: Is 12 bytes worth for Timestamps?

Another question, if I turn on Timestamps, what about TcpInitialRTT? (the value is decribed on http://www.microsoft.com/windows2000/te ... /58802.asp)

By the way, I found large DefaultReceiveWindow and DefaultSendWindow would decrease Audio and Video qualities of MSN Messenger and Netmeeting. I am from Canada, and I talk with my mom from China everyday on MSN Messenger.

For Roger Cable(5000/768), I use 65536 for DefaultReceiveWindow, and 8192 for DefaultSendWindow.

mnosteele52 · Post by **mnosteele52** » Sun Dec 28, 2003 10:55 am

Good point but you answered your own question..... sort of. The 12 bytes it adds are not worth enabling it.

Try and see for yourself, I have experimented with it and found it to have negative effects.

Something you might try is what I have posted in the poll I started.

In CableNut make the following changes:

GlobalMaxTcpWindowSize = leave blank
Tcp1323Opts = 1
TcpMaxDupAcks = 2
TcpWindowSize = leave blank

See if those changes improve your connection.

It's nice to see someone take the time to read the White Papers and other literature and question and try new things.

wangnan · Post by **wangnan** » Sun Dec 28, 2003 11:52 am

ok!

One tip for testing:
the result will not be obvious if you test just one thread.
try to do 10 speed tests at same time, and add the values together.

You can download a free multi-threads download manager called NetTransport and try to use it to download a 100mb file with 10 threads. this is the best tool for testing internet performance.
http://www.nettransport.com.cn/default.htm
(there's some useful configurations inside, read Help)

By the way, did you try timestamps testing with a terminal on another side of earth?

mnosteele52 · Post by **mnosteele52** » Sun Dec 28, 2003 12:39 pm

That is pretty much just a download manager. I tested with various FTP tests in different locations.