[242] in Info-AFS_Redistribution
rs_aix31 cache strangeness
daemon@ATHENA.MIT.EDU (David Brelsford)
Tue Jul 23 16:22:01 1991
Date: Tue, 23 Jul 91 15:31:30 GMT
From: brels4d@minnie.nic.kingston.ibm.com (David Brelsford)
To: info-afs@transarc.com
Reference: Append from paw@rigel.Dartmouth.EDU, 12 Jul 91
I've also had problems with caching on RS/6000s. Following the
Sysadmin Guide recommendation, we set our cachsize to about 95%
of the partition dedicated to the cache. With a partition
size of 32meg, the 95% rule seems to work. I was able to successfully
copy a 45meg file from an afs directory to local without incident.
With smaller caches we've seen the same kinds of problems as Pat.
I experimented a bit today and discovered that with a 24meg partition,
I could not set the cachesize to 90% without seeing
the cache hang up when trying to copy a 30meg file from afs to local.
I get an i/o error from the copy, then am unable to execute binaries
located in afs directories. Pat's trick of decreasing the cachesize
then increasing it makes the binaries in afs usable again.
It was similar when using a 12meg partition for the cache, copying
the 30meg file failed when the cachesize was set to 90% of the partition
size, but succeeded when cachesize was 85%.
Apparantly the 95% rule doesn't apply all of the time. Are there
guidelines for setting cachesizes which work? Better yet, can
we expect the cache manager to recognize and recover when there isn't
enough space in the partition containing the cache?
Dave
