Re: tstile in less than a day

[Sad Clouds <cryintothebluesky%gmail.com@localhost>]

On Sun, 10 Mar 2019 15:44:49 +0100
Martin Husemann <martin%duskware.de@localhost> wrote:

> On Sun, Mar 10, 2019 at 02:41:32PM +0000, Sad Clouds wrote:
> > In the past I found NFS on NetBSD quite flaky, i.e. I/O hanging
> > indefinitely under high load, not just sparc64, but amd64 too.
> 
> I have seen that a lot on all architectures in the past, on both sides
> (client and server), and usually it was some driver bug that causes
> packets to be deterministically lost.
> 
> But now that tcp mounts are the default, this should be handled
> better. This needs closer inspection by packetraces on both sides.
> 
> Martin

When you say "some driver bug" are you referring to network card driver
or something else, like NFS subsystem? I always use NFS with TCP and
any packets lost should be re-transmitted by TCP. I suspected bugs in
NFS, since I could always reliably reproduce it by setting NFS rsize
and wsize to something like 65536 and then start copying large amount
of data. Sooner or later NFS would reliably hang.