[OpenAFS] client problems

Martin Schulz schulz@iwrmm.math.uni-karlsruhe.de
10 Apr 2001 09:48:05 +0200


I am on the way to install openafs on our (mostly
Linux-)pool. However, I experience serious problems with the client
side. The server seems to run ok, but the client shows ridiculous performance, 
most of the time it is just sitting there doing nothing (as it seems).

The behavious is not consistent, that is only reproducable to a
certain degree. Yesterday, I moved a 1MB File from a local disk to the
afs server:

$ time mv /tmp/bla /afs/iwrmm.uni-karlsruhe.de/user/schulz/
real	3m55.986s
user	0m0.000s
sys	2m19.870s

Where I could observer 5 afsd processes running at full tilt,
constantly using approx 39,7% CPU each, whereas on the server,
everthing was calm.

After a afs stop/start on the client machine this morning, I now can observe
another (by no means better) behaviour. The read performance seems to be ok,
but the writes seem to run into a timeout:

$ time cp /afs/iwrmm.uni-karlsruhe.de/user/schulz/blub /tmp/blub

real	0m0.016s
user	0m0.010s
sys	0m0.010s

$ time cp /tmp/blub /afs/iwrmm.uni-karlsruhe.de/user/schulz/blub          

real	3m20.605s
user	0m0.000s
sys	0m0.030s

The client machine is a dual pentium box, running 2.2.26-22 kernel
(RH7.0), using openafs 1.0.3 built from the src.rpm. The prebuilt
stuff did not work for me (maybe I did not wait long enough..)

I suppose some timeout happening here. Are there some known locking
issues with dual processor configurations? How can I get deeper
diagnostic information on what afsd is trying to do?

Starting afsd with strace did not yield any suspect results...

Martin Schulz                             schulz@iwrmm.math.uni-karlsruhe.de
Uni Karlsruhe, Institut f. wissenschaftliches Rechnen u. math. Modellbildung
Engesser Str. 6, 76128 Karlsruhe