Dear Jef,
Thanks for your response.
Win98 gave bad results as ppp did not allow mtu's above 576.
Win95 works fine untill 400kbyte/s if winsock2 is used, but the cpu should be absolutely free (do not touch the mouse, let DVB
decoding be done by a NT).
NT allows max of 160kbyte/s.
(all machines 233 pentiums)
Fully using the window was not possible as losing a few packets was enough to degrade performance heavily (performance is more
depending on terestrial network and pc performance, we were not able to fill windows larger then 0.5mb without instability).
We used 19k2 phoneline's and did not see choking with ack's as 10mb network, routers and cpu's were having more impact.
(increasing to 100mb or 155mb network and using Solaris or Linux could stretch performance better)
(NT is having also a good stack but only 64kb windows, let's hope Bill will implement large windows soon!!)
(running on low latency networks we have seen TCP stack performance as win95:NT:Linux=1:3:6)
Conclusion:
Nice to have a lab environment doing high throughput.
But no enduser will ever see these figures as in the real world you will lose a few packets!
We will need better rfc's to handle the difference between congestion and losing a few packets!!
Jack Stiekema
Cap Gemini Nederland
Phone: +31 (0)30 252 9219
Fax: +31 (0)598 398008
-----Original Message-----
From: Harrang, Jeffrey P <[email protected]>
To: '[email protected]' <[email protected]>
Date: dinsdag 1 december 1998 19:57
Subject: Your question about TCP/IP tuning
Hi,
Noticed your thread on increasing the TCP window size on NT/Win95 platforms.
Two questions:
1) Were you successful in getting the window size increased and did you see
the expected performance increase?
2) You mention using a telephone line for the return traffic. What rate was
that at and how did you resolve choking it with ACKs?
Thanks,
Jeff
This archive was generated by hypermail 2b29 : Mon Feb 14 2000 - 16:14:51 EST