I got a quick response from Casper Dik, the solution is
increasing the tcp xmit/recv buffer size on SunOS 4.x boxes
from the default 4k to 8k. This means changing the 2
variables tcp_sendspace & tcp_recvspace in
/sys/netinet/in_proto.c, rebuild kernel, install and reboot.
I tested the change on a expendable box and the command run time
went from 25m24s to 6m35s. This test in this case involves
different subnets so timing is not as repeatable but the
improvement is obvious.
Thanks to Casper, Michael Kriss, Vinnie German, Paul Kloepfer
Casper Dik wrote:
> YEs, increase the TCP buffer sizes on the sunOS 4.x boxes;
> this requires a kernel reconfig.
> The problem is that the Solaris boxes offer a window that is much
> bigger than teh SunOS 4.x boxes can accept; so they never fill teh window
> for more than 1/3. SOlaris delays acks for longer than that so it delays
> for 200ms every 4K received (that's when Solaris sends the ack)
My original question:
> We're running remote backup of SunOS 4.1.4 machines using
> rsh like this
> rsh sunos4 -l backupid /usr/etc/dump 0f - /dev/rsdxx
> There's a major time difference when this command is run
> on SunOS 5.6 vs. SunOS 4.1.4, the 5.6 backup server takes
> about twice as long given the same backup client & filesystem.
> I've tried this on 3 different 5.6 boxes (2 at 105181-12 kernel
> 1 at 105181-14) with 3 different 4.1.4 boxes. All machines
> are on the same subnet connected to different hubs, that
> should eliminate network equipments as the problem cause.
> Some test runs with rcp /vmunix also shows the 5.6 boxes to be
> slower although the difference is not as much.
> Any suggestions on what can fix this problem?
Viet Hoang email@example.com
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:14:08 CDT