On Fri, Sep 02, 2005 at 11:25:08AM +0300, Mike M. Volokhov wrote:
> Seems you're right. On NFS server we have installed extremely cheap
> Realtek 8139 based NICs :-/
> 
> With 8192 sized data I see periodically burst speedups, and the rest
> time my NFS client is locked. With 1024 I haven't obscure this kind of
> problems.
> 
> But with default settings (32k?) my clients just infinitively locked
> for hours (no CPU load, no network traffic). Is this a normal behaviour,
> even with such bad NICs as 8139?

I think you should try TCP mounts, then. The TCP flow control may work around
the issue.
The problem here is that, with UDP mounts (the default), if a packet gets
lost, the system will send again the exact same 32k stream of packets,
which will get lost again, because the NIC on the other end can't handle
a burst of 32k. In such a situation TCP would decrease the window size and
the retransmit would not overlow the receiver's buffers.

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
     NetBSD: 26 ans d'experience feront toujours la difference
--