In article <u93ehsd2mm.fsf at wrasse.wustl.edu>,
Sean Eddy <eddy at wrasse.wustl.edu> wrote:
>Arne Elofsson <arne at elof.biokemi.su.se> writes:
>> No problems, however NFS prestanda of linux is really bad (sometimes a
>> factor of 4 slower than on solaris). So write to local disks when you
>> write a lot of data
>>Not sure what you mean by "prestanda" here.
>>Linux NFS is just fine, but comes configured in a way that typically
>takes a 4x speed hit. The default NFS read/write size in Linux is 1024
>for historical reasons. For better performance on disks mounted from
>Solaris or IRIX NFS servers, set wsize=4096,rsize=4096 in the options
>in /etc/fstab for NFS mounted disks.
Yes, Linux NFS *client* performance is fine if you do this, but the nature
of its server means that its server performance is dire. This, as another
poster has said, will get a lot better in 2.1.x kernels, where at last the
NFS server has been moved into the kernel, where it should be (and indeed
where it is in most commercial UNIX implementations, including IRIX)
Tim.
