> "Steven N. Hirsch" <shirsch@adelphia.net> writes:
>
> > Alan,
> >
> > Yes, it does go rather fast. Unfortunately, where it leads me is into
> > oblivion. Running 'iozone 64 8192' on a dual-PPro SMP client (2.2.1-ac3)
> > against a knfsd server locks up solid after a few seconds. No log
> > messages, no fanfare.
> >
>
> Hm, very strange. I don't have an SMP machine here, but I must admit
> that Trond's patch (in it's v0.9 incarnation) runs rock solid here.
>
> Also, I have a feeling that Trond's code is the best thing that
> happened to Linux NFS, ever!
>
> In 2.0 era, I was writing at 20-50KB/sec (sic!). When knfs got
> introduced in 2.1, performance slightly raised but never above
> ~300KB/sec. And that data comes from 100MBps fdx connection!, between
> fast machines, unloaded network.
Hmm. What is the architecture of the server you are running this against?
I see ~3.5-4 MB/sec. write and 2-3 MB/sec. reads between two Pentium class
boxes running 2.2.1. This is with a moderately congested network using a
non-switched hub (i.e. half-duplex).
> Now, I'm spending last few days constantly writing and writing at
> ~1MB/sec. Unbelievable. It's becoming addictive. :) If only I could
> utilize wsize of 32768, I bet it would be even faster (currently, code
> allows me to raise wsize up to 16384, and I can't find where the limit
> is).
Preliminary benchmarks suggest that the new write-gathering is
considerably _slower_ between the same two boxes. More later.
Steve
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/