| This looks very similar to the ipmt (C961205-932) that I have been working
for a long time. We do not have a fix (the ipmt is still attempting to collect
data on the problem for CA) but do know what seems to trigger it;
Very heavy io load (normally backup is running) with hotfiles data collection
enabled. The psdc$dc_server goes compute bound and no data is written out (or
the data is garbage- no reports can be run off the data). The collector will
normally recover after a while (when the io load decreases??) and data will be
written. While it is in the compute state, advise coll/stop/wait may not work
but $stop psdc$dc_server should.
workarounds that reduce the likelyhood of the collector going compute bound;
turn off hotfiles data collection completely (not acceptable to many customers)
change the hotfiles limit from .33 to 1.00 (only show real bad disks' hotfiles)
turn off hotfiles data collection during backup and turn back on after backup
reduce the number of concurrent backup/production jobs to reduce io load.
(least acceptable to customers)
|