Subject: interfaces, receive buffers, and NFS
To: None <>
From: Frank van der Linden <>
List: tech-net
Date: 04/02/2001 14:01:34
Because a lot of older NICs were apparently unable to handle
32k UDP datagrams (i.e. a quick succession of packets), I
set the default read and write size back to 8k. This value
can be overridden in <machine/param.h> or a config file.

For the i386, I set it to 32k. I know that some old ISA cards
might have trouble with this, but that problem has
always existed, even for 8k, and setting the value to the lowest
common denominator would land us at a value of about 2k, which
gives horrible performance (the bump from 8k to 32k can
make a 30% difference in throughput). And, for those cases,
a mount parameter change or a kernel option override will
work just fine.

Anyway.. what would really be useful is to have interfaces export
their receive buffer size, so that upper layers can do something
useful with that value. Limit the rsize and wsize in the case
of NFS. Would it be a good idea to add such information to
the if_data structure?

- Frank

Frank van der Linden                 
Quality NetBSD CDs, Support & Service.