Re: NFS client performance 1.5 orders of magnitude too slow? (fwd)

Godmar Back (gback@cs.utah.edu)
Tue, 9 Mar 1999 17:22:20 -0700 (MST)


>
> > For 2.2.3 (without the 2.2.2ac7 patches), a wsize of 8k appears to
> > be slightly faster than a wsize of 4k when writing to a BSD ffs (*) hosted
> > on an i386 with a 4K page size. (30.5 vs. 31.5 seconds for a 12 MB file)
> >
> > Is this because 2.2.3 is broken compared to 2.2.7ac1 or is this
> > contradicting your statement that it only matters for BSD+8K pagesize
> > machines?
>
> Thats probably just luck or better client side throughput through less
> syscalls

It was with the same client; I did "cp bigfile > /dev/null" to prime
the cache (does that prime the buffer cache?) and then
"time cp bigfile server". I'd expect the same number of syscalls
when I do that.

>
> 2.2.3 is "broken" if you like for the 8K case yes. 2.2.2ac7 isnt, but we are
> still combining the two to get the bugs out of both so we have a final good
> NFS client
>

Hey, I'm all for it. I think it's great.
I have no desire whatsoever to go back to the times where it was faster
for me to ftp a file (including typing my password and cd in the right
directory) then to copy it via NFS (especially for those .no4 files.)

I'll monitor the next release(s) to see what comes out of it.

- Godmar

-
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/