unknown
2006-05-15 16:07:55 UTC
I'm getting a strange problem where loading of 200Mb tran
logs can take 4 minutes on one run and 30 minutes on
another. The problem is occurring between the analysis and
redo phases of the process.
When things are running as normal, there is only a few secs
delay between the completed analysis message and the start
of the redo phase messages. However when the process slows,
the gap between these can be over 20 mins, while the server
process performs continuous phys_io. There are no other
server processes, other than backround tasks, and the same
logs can take vastly different times to load. There is no
pattern to it and once it occurs subsequent logs can run
just as badly, although after time they can start showing
good performance again.
Is there anything in the way the logs get applied that can
explain the inconsistent delays. We're not necessarily
blaming the ASE for the delay but we can't currently see any
reason for it at an OS level. I was just wondering if anyone
else has ever had a similar problem.
John
logs can take 4 minutes on one run and 30 minutes on
another. The problem is occurring between the analysis and
redo phases of the process.
When things are running as normal, there is only a few secs
delay between the completed analysis message and the start
of the redo phase messages. However when the process slows,
the gap between these can be over 20 mins, while the server
process performs continuous phys_io. There are no other
server processes, other than backround tasks, and the same
logs can take vastly different times to load. There is no
pattern to it and once it occurs subsequent logs can run
just as badly, although after time they can start showing
good performance again.
Is there anything in the way the logs get applied that can
explain the inconsistent delays. We're not necessarily
blaming the ASE for the delay but we can't currently see any
reason for it at an OS level. I was just wondering if anyone
else has ever had a similar problem.
John