RE: windows 95 and NT

From: Jack Stiekema ([email protected])
Date: Mon Aug 10 1998 - 15:25:57 EDT


> -----Original Message-----
> From: Tom Saam [mailto:[email protected]]
> Sent: Monday, August 10, 1998 1:15 PM
> To: Jack Stiekema
> Subject: RE: windows 95 and NT
>
>
> Jack-
>
> Two things:
>
> 1) TCP Slow Start (windowing) hurts small file sizes or
> links with any kind of error rate.

Thanks Tom,

Tracing gives a nice 1, 2 ,3 4, 5, 6 packets, and then it't stable 6 (5x1500
and 1x932)so i guess the NT does not like to increase it's default window
size despite of all the papers who say it should.
I also had the suggestion that maybe the NT just does not have enough send
buffers

I have a Philips collegue visiting Microsoft in the US this moment, i just
mailed him my questions let's hope he get's the answers.
The question he will ask Microsoft:
"
Server is NT4 (build 1381,sp3, winsock2 entry in registry),
client is NT4 (build 1381,sp1, winsock2 entry in registry),
in between i have a 200ms WAN/satellite simulator.
So the total RTT is 400ms on a 10mb ethernet link.
I made on both NT's the the MTU 1500, the TcpWindowSize 14800, enabled MTU
detect and disabled black hole detect but:

If i make a ethernet trace (of a ftp session, and of a tcp test tool) i see
that there are 5 packets of 1500 and 1 of 932 bytes, this equals 8432 the
default window size of Microsoft.
As there is only 8kb every 400mS the total transfer rate is 22kB a second,
and not the 10meg we were hoping.

Whatever i do (use a win95 winsock2 client, put windowsize in all
device/parameter/tcp registry entries) i can not get the window size larger
then 8k.
I need to enlarge the RWIN to 1meg in order have decent transfer rates at
satellite links

So the questions is:
How tho enlarge the TcpWindowSize to 1 meg on win95 and NT4?
"

Jack Stiekema

Cap Gemini Nederland bv
Industry, Product RElated Software Solutions
Phone: +31 6 22240066
Fax: +31 598 398008

>
> 2) TCP typically generates 2 acks for every MTU, so the
> dial return
>
> ack rate can be a bottle neck.
The second one should be ignored by the receiver, so it should not hurt.

>
> ....just some ideas!
Thanks for the help Tom and all the others.





This archive was generated by hypermail 2b29 : Mon Feb 14 2000 - 16:14:46 EST