Subject: Re: OT - Mac network diagnostics?
From: Tom Lane (tgl@sss.pgh.pa.us)
Date: Fri Jul 20 2001 - 19:38:24 EDT
Lance Levsen <l.levsen@printwest.com> writes:
> The scenario is:
> 100baseT network,
> Linux PostgreSQL database
> Mac RealBasic application connects to the database
> Small queries happen fast, Large Macintosh LAN queries take
> forever, 6-13 seconds. W2K and an Ultra Sparc clients get
> returns of .08 seconds w/ the same query.
> The strange thing is that the same query from a remote Mac
> client via Internet takes about 1.5 sec.
That makes it a networking problem, not a Postgres problem, so
it's a bit out of my depth. However, I did just yesterday discover
a hack that helped me with a Linux networking problem: it seems the
Linux kernel has some weird retransmit behavior when there is a lot
of data pending in the send-queue. Maybe you are being bit by the
same thing. Try doing this on the Linux box:
echo 0 > /proc/sys/net/ipv4/tcp_retrans_collapse
and see if things work any differently. This parameter is documented
thus:
tcp_retrans_collapse - BOOLEAN
Bug-to-bug compatibility with some broken printers.
On retransmit try to send bigger packets to work around bugs in
certain TCP stacks.
and I'm currently thinking that it creates more bugs than it fixes...
but it seems to be defaulted ON.
regards, tom lane
This archive was generated by hypermail 2b28 : Sun Oct 14 2001 - 03:04:45 EDT