We've been noticing a performance problem with afs 3.3 and 3.3a
on pmax_ul43a (both ultrix 4.3a and 4.4), and I'd like to hear from
anyone else that is experiencing it.
The symptoms we're seeing are:
1. Sometimes there is a pause of about 10-20 seconds when
running small programs out of afs that have been run recently.
There is no disk activity during the pause and the process
is not interruptible. Other disk accesses seem to block as
well until the program in question comes back. Eventually
the program starts and the machine generally seems to come
back to life.
2. The time it takes to stat() large numbers of files is greatly
increased. We use the "depot" program from Andrew, which spends almost
all of its time stat()ing files in afs, and the running time
of this program changes from 8 minutes to about 50 minutes
when all I change is the version of afs from 3.3beta to 3.3 or 3.3a.
Afsd is being run with "-nosettime" but no other arguments.
I realize that these things happen from time to time with afs in
general, but the performance difference between 3.3beta and 3.3 or 3.3a
is enough that the machine seems quite a bit slower in general.
Anyone seen anything like this?
(This is with the binary distributions from Transarc.)
-Chuck Silvers, CMU School of Computer Science Facilities
