Subject: Re: NFS problems
To: David Laight <firstname.lastname@example.org>
From: Jukka Marin <email@example.com>
Date: 07/20/2002 01:32:29
On Fri, Jul 19, 2002 at 11:13:50PM +0100, David Laight wrote:
> > It's a 16-port switch with an 8K-entry MAC cache and a 512 KB packet
> > buffer. Shouldn't have any problems routing the packets between any
> > two machines in my home network. ;-)
> > Besides, as far as I can tell, the problems started when I started
> > upgrading the client systems from something like 1.5W to 1.6beta or
> > so. No hardware or network changes were made.
> Maybe someone has improved the netbsd code so that it now sends out
> the packets closer together - so the switch can no longer cope.
> Also it (probably) allocates buffer space to each port. 512k/16
> is 32k per port. A little bit over overhead and it will all
> go horribly wrong.
Can't believe a new switch couldn't cope with normal IP traffic..
Here's an URL: http://www.computex.com.tw/archive-detail_new.asp?index=3044
At least they make it _sound_ like it should work just fine (especially
with only a few ports in use).
I don't know how switches manage buffer space, but I hope they're not
stupid enough to allocate 32 KB for ports that are not connected at all..
Besides #N, my server is still running 1.5R and I haven't even rebooted
it for ages (204 days). I can't see why it would suddenly start sending
out data much faster than before (even though Jason has optimized the
TCP/IP stack recently :-)
It sounds more like a new bottleneck in the new NetBSD versions ran by