On Fri, Dec 21, 2012 at 02:34:50PM +0100, Jan Stary wrote: > On 5.2/i386, man sysctl says > > To adjust the number of kernel nfsio threads used to service asynchronous > I/O requests on an NFS client machine: > > # sysctl vfs.nfs.iothreads=4 > > The default is 4; 20 is the maximum. See nfssvc(2) and nfsd(8) for > further discussion. > > Does it still apply? The default now seems to be > > # sysctl vfs.nfs.iothreads > vfs.nfs.iothreads=-1 > > which scales itself to 4 when I copy a file from a NFS server. > Should I be touching vfs.nfs.iothreads manually? > Setting it to 20 just panicked my 5.2/i386. > > Also, neither nfssvc(2) nor nfsd(8) seems > to contain any "further discussion" of this. >
i don;t use nfs, so a bit of guesswork... i get -1 here too. but you;re saying that the value changes to 4 when transferring, right? i can;t see anything from a quick glance at cvs logs that indicate past discussion of the sysctl. but i take it that nfsd's -n option (default 4, max 20) is analogous, and that the stuff about how many daemons to run is what the "see nfsd..." reference is to. and that the default is 4 when nfsd is actually doing something. so it sounds like the docs are correct, but not very clear (or helpful). if some nfs dude can clarify, we can tidy it up a bit... jmc
