tech-kern archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: how can the nfs timer work?

On Mon Jan 26 2009 at 13:17:37 +0000, Andrew Doran wrote:
> A workaround appropriate for 5.0 would be to take softnet_lock around the
> entire loop, and comment out the solock/sounlock calls. This would lock all
> inet/inet6 sockets.

Sounds good, although ISTR from a recent discussion on this list that
NFS over ISO was considered important.

In other news, I was happily using rump_nfs with the hack I posted for
quite a while, but it seems to have lost its magic.  I wonder if this
coincides with some kernel changes (I run rump file servers usually from
the very bleeding edge) or me starting to run 5.0 instead of 4.99.42 on
my host.  Actually, it seems that both the kernel and rump nfs clients
have poor performance on 5.0.

Anyway, just reporting some observations, haven't really had time to
think about them more carefully or dig into them, so the problem might
be somewhere completely different, e.g. network or the server.

Home | Main Index | Thread Index | Old Index