sjg@quick.com.au (Simon J. Gerraty) wrote:

>> I just hung my 1.2G box... by writing too quick to an NFS server.

>> Client is P166 running the recent 1.2G snapshot, with 3C509 nic.
>> Server is i486/DX33 running a 1.2F kernel (Jun08) and a WD8003 8bit
>> NIC.

Dave Burgess <burgess@cynjut.neonramp.com> wrote:

>Mine did the same things, except the server is a P100 and the client is
>a 386/40.  The error message and the network card are (oddly) precisely
>the same.

>> Jul 12 22:49:40 zen /netbsd: ed0: warning - receiver ring buffer overrun
>> Jul 12 22:49:40 zen /netbsd: ed0: warning - receiver ring buffer overrun

>The system with the 8003 is wedged on my end.  Same symptoms as yours.

>> I'll remount with write size set at 1k... after I reboot.

>Since I'm seeing exactly the same problem, but on the slow machine, I
>doubt it is actually NFS.  My guess would be a recent change to the 8003
>code, since that is actually the only thing the two set-ups seem to have
>in common.

	Your problems are that the wd8003 card simply *cannot* handle the
8k packet sizes (or if they can handle 8k packets, they can't in large
volume).  The solution is to mount all NFS mounts on the machine
with the 8003 with both 1k read and 1k write block sizes.

	This is not a change in the ed* code -- the 8003 never has worked
with 8k packets, at least not in the three years or so that I've used them.
I remember over two years ago trying to make an nfs-boot kernel and needing
some kernel options for limiting nfs block sizes before it would boot.

Cheers,
Phil
--
Phillip F Knaack
Systems Administrator, Information Development for Extension Audiences (IDEA)
Iowa State University Extension