[2346] in linux-net channel archive
Slow PPP & 3c509 vote re-issue.
daemon@ATHENA.MIT.EDU (Kristofer Karas)
Thu Apr 4 08:44:04 1996
Date: Wed, 03 Apr 1996 15:41:50 -0500
From: Kristofer Karas <ktk@enterprise.bih.harvard.edu>
To: Linus Torvalds <torvalds@cs.helsinki.fi>
CC: linux-net@vger.rutgers.edu
Linus Torvalds wrote:
> On 1 Apr 1996, Christoph Lameter wrote:
> > I think I isolated the problem. Alan changed the ACK handling in 1.3.73. Ever since then
> > we have the performance problems because Packets are resent and resent.
> How isolated have you gotten it? Can you check if it's the tcp.c or
> tcp_input.c changes?
Linus (et al),
This assertion by Chris Lameter is supported in that it affects Ethernet performance, too; to
recap the benchmarks I noted here a few weeks ago w.r.t the v1.0.7 3c509 driver:
Two Pentium-90 linux systems with 3c509 cards directly connected on 10baseT net,
Linux 1.2.13 talking with 1.2.13: 600+ KB/sec.
Linux 1.3.75 talking with 1.2.13: 400 KB/sec.
Both these machines are used in a hospital environment (one's a hot-swappable backup to guarantee
clinical system integrity 24 hours/day) with up to 100 users each; minus the pomposity, this is
just saying their network load is heavy. :-) I presume my vote for the patched v1.0.7 3c509
driver fell on the floor, so let me re-issue it here: The v1.0.3 driver in the current kernel
does cause network lockups when back-to-back packets are handled; the v1.0.7 driver has shown
itself to be stable in this heavily used environment, thus has my blessing. (Alan take note.)
-- Kris
