[OpenAFS] OpenAFS speed - some benchmarks

[Derrick J Brashear]

On Tue, 24 Jun 2003, Andrei Maslennikov wrote:

> Current results may be found at the following URL:
>
>    http://afs.caspur.it/slab/CASPUR-Slab-0603.pdf
>
> (Skip to page range 22-25, other tests have nothing to do with AFS).
>
> I would love to be able to report the better numbers, but AFS
> performance does not look well for large files (and we used a
> large cache on the ramdisk, and played with chunksizes and/or
> fileserver params...). NFS is simply doing 2.5 times faster..

That is pretty sad, you're right. I think some work needs to be done to
determine which of:
-fileserver
--just inode
--just namei

-rx
--with jumbograms as is
--with jumbograms disabled
--with older 3.4 style jumbograms

-client

is really the source of the suck. i suspect it might be possible to
improve write performance by parallelizing cache and network work, but
really I haven't poked enough to know "how"; The question is, where can
the most gain be had for the least work.