On Jan 03 10:06:38, [email protected] wrote:
> On Sun, Dec 23, 2012 at 08:01:46PM +0001, Jason McIntyre wrote:
> > On Sun, Dec 23, 2012 at 08:17:01PM +0100, Jan Stary wrote:
> > >
> > > Yes, on the _client_ (sorry, I should have been explicit).
> > > What got me trying it is that sysctl(8) says
> > >
> > > To adjust the number of kernel nfsio threads
> > > used to service asynchronous I/O requests on
> > > an NFS _client_ machine.
> > >
> > > Your diff seems to imply that this affects the _server_.
> > >
> > > However, on the client, vfs.nfs.iothreads jumps to 4
> > > as soon as I mount the NFS share. On the server, it stays at -1.
> > >
> > > On the client, when I start copying from the server, it stays at 4;
> > > on the server it stays at -1. Same when I copy the other way round.
> > >
> > > So it really seems to affect the client (as sysctl(8) currently says),
> > > which I think makes your diff incorrect.
> > >
> > > After I unmount the share on the client, vfs.nfs.iothreads
> > > stays on 4; I set that manually to -1, and it jumps to 20,
> > > without even having anything NFS-mounted. Is that intended?
> > > Setting it manually to anything but -1 then panicked my
> > > 5.2/i386 client. (I would try current, but that's currently
> > > not bootable on my i386 laptop.)
> > >
> >
> > ok, so unless someone steps up and explains to me how this is meant to
> > work i can;t do much...
> >
>
> so after some help from blambert, i've committed a doc fix.
Thanks, less confusion.
vfs.nfs.iothreads is still mentioned (and commented out)
in the default /etc/sysctl.conf - should that go too?
> i've not
> documented the fact that -1 is kind of the default. i'm not sure if
> explaining it is that helpful. nor do i know why it's done this way.
Could a NFS dude please kindly shed some light on how
vfs.nfs.iothreads works? In particular, is it supposed
to affect the client or server behaviour?
If the user is supposed to be able to use it,
the manpage should describe how it works, right?
If OTOH the user is not supposed to touch vfs.nfs.iothreads,
the documentation should clearly say so (and it probably
would have to go from sysctl.conf). I for one can happily
leave it to the OS, and only tried bumping it up on my
busy NFS client upon reading about it in sysctl(3,8).
Thank you
Jan
