[OpenAFS] Extremely poor write performance.

[Rubino Geiß]

Hi Jimmy,

> Robin Yamaguchi <rhy@physics.ucsb.edu> writes:#
> > Here are some of the results from running bonnie++ (HD/filesystem
> > benchmarker) on an afs directory:
> > 
> >      ------Sequential Output------ --Sequential Input- --Random-
> >      -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- 
> --Seeks-- Size 
> > K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
> > 496M  2351  24  2516   9   772   2  2793  25  3480   2  43.6   0
> >      ------Sequential Create------ --------Random Create--------
> >      -Create-- --Read--- -Delete-- -Create-- --Read--- 
> -Delete-- files  
> > /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP /sec %CP
> >   16   159  26  2013  82   198  18   162  28  1013  70   215  15
> 
> I dont think bonnie is a relevant benchmark for network filesystem.

Why not? I know there are reasons, but its not invalidation the big picture,
does it?

> 
> Have you tried something like:
> 
> 1) Extract a tarball
> 2) compile a program
> 

As I already posted this is what we do on daily basis. It's our work, so to
say _and_ make clean does take a _lot_ of time. Say: 2-20 times slower than
NFS and 5-500 times slower than local ext3!!! (exact figures depend on
server load and the exact program you are building/cleaning)

> Or maybe you should try the andrew benchmark, which is more 
> relvant but very old.

I check it.

> 
> However if your are intressted in afs-performace-numbers you 
> can compare with mine, statistics are done with afsfsparf 
> from the arla-package. This is relevant when you test 
> fileserver-performance, it does not performancetests 
> client-performace.
> /Jimmy

I was not able to build it, up to now: missing libraries, strange config
options, etc.

Bye, Ruby