Re: Actual window sized used
It has been a couple of years since I tested this but creating the registry
GlobalMaxTcpWindowSize used to set a hard limit while the application
could exceed TcpWindowSize. [Pointers to documentation in
http://rdweb.cns.vt.edu/public/notes/win2k-tcpip.htm might be of
interest, e.g. if you haven't already seen the Microsoft TCP whitepaper
(it includes the details of the OS making TCP window a multiple of MSS).]
Anyway, I'd be tempted to increase the value of GlobalMaxTcpWindowSize and
try that too. (And then tcptrace on a packet capture just to be sure.)
Paul Hyder
NOAA Forecast Systems Lab
Boulder, CO
Chris McFarling wrote:
>>Do TCP window parameter settings in the OS affect the window size used
>>by
>>iperf? For example, if I make the following registry edits in Win2K:
>>
>> GlobalMaxTcpWindowSize=17520
>> Tcp1323Opts=1
>> TcpWindowSize=17520
>>
>>and the start iperf with the following:
>>
>>iperf -s -w 65700
>>
>>
>>will iperf use a 65700 byte window or will it use a 17520 window? It
>>reports
>>that is using a window size of 64K but I just want to verify that what
>>it's
>>reporting is actually true.
>>
>>Thanks
>>
>>
>>Whatever window size you set on iperf's command line will override the
>>OS setting. If you don't specify a window using -w then iperf will use
>>the window size your OS specifies.
>>
>>--Andy
>
>
>
> Taking that a step further, I've noticed the requested and actual window
> sizes are always the exact same. For example, on OS X when I request a
> window size of 17520 iperf reports:
>
> TCP window size: 18.4 KByte (WARNING: requested 17.1 KByte)
>
> Does iperf automatically choose a window size that is a multiple of the MSS?
> In the case of OS X the MSS is 1448. 17520 is a multiple of 1460. The closet
> multiple of 1448 (rounding up) would be 18824, or 18.4K.
>
> Just looking for verification on that.
>
> Thanks
>