Re: Very poor NFS I/O

[Ignatios Souvatzis <ignatios%cs.uni-bonn.de@localhost>]

On Tue, Mar 03, 2026 at 07:41:55PM +0100, Hauke Fath wrote:
> On 3/1/26 00:17, MJ wrote:
> > Timeout settings due to soft mounts? Read/write buffers too small?
> > I generally do hard client mounts with rsize/wsize set to 32K.
> 
> That would be for udp nfs mounts. IIRC, default is tcp.

On ancient *client* hardware with small-buffered NE2000-alike NICs,
using UDP mounts:

you had to mount with rsize=1024, to ensure that no back-to-back IP
fragments where sent by the server, else one of them would surely be
lost by too slow interupt handling, and the RPC layer would retry the
whole NFS request, creating another 6 fragments for rsize=8192, possibly
at infinitum.

Note: the server would do some superfluous work, but not be clogged down
too much. The client would not proceed, or only slowly.

But I hope this is not the issue for the clients in question.

	-is
-- 
Ignatios Souvatzis, Chief IPv6 enabler		RFC 6540
Gemeinsame Systemgruppe b-it + Informatik	Tel. +49 228 73-60701
gsg%cs.uni-bonn.de@localhost