lkml.org 
[lkml]   [1998]   [Mar]   [11]   [last100]   RSS Feed
Views: [more markup]  [less markup]  [headers]  [forward] 
 
Messages in this thread
/
DateWed, 11 Mar 1998 00:22:33 -0800
From"David S. Miller" <>
SubjectRe: pre-2.1.90..
   Date: 	Wed, 11 Mar 1998 07:07:29 +0000 (GMT)
   From: Matthew Kirkwood <weejock@ferret.lmh.ox.ac.uk>
   Over relatively quiet 10mb/s network, I can get no more than 830k/s
   ftp, where 2.1.8x were doing only about 720k/s.  2.0 and previous
   2.1 would often get upwards of 1mb/s.

   Local throughput is a lot better -- up to 7.5mb/s from about 3mb/s
   on .89; there's always a stall after about the first 25k, though --
   is this related to the bandwidth problem?

I have things cranking a bit more at the moment.  BTW how are you
measuring localhost thruput?  If with lmbench, run it a few times
before you come to numerical conclusions because:

1) 1.[01] lmbench's bw_tcp did a quick write and could be suspect
   to general noise during connection startup.  Best bet here
   is to run it a few times, or give it a buffer size arg such
   as "bw_tcp localhost 20m"

2) 2.0-alpha lmbench's bw_tcp acts a bit hokey to me, it tries to
   increase the buffer size until the noise goes away.  The buffer
   size increment scheme is not always fool-proof.  Sometimes it
   chooses a buffer size, that with the sockets send buffer size,
   will cause a very tiny send packet to be generated.  So you'd
   see on the wire:

     full sized frame
     full sized frame
     frame with 32 bytes

   or something like this.  I've had good luck with again giving
   it the "20m" arg and running it a few times.

Later,
David S. Miller
davem@dm.cobaltmicro.com

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu

\
 
 \ /
  Last update: 2005-03-22 12:41    [from the cache]
©2003-2008