Checkpoints can take too long on large shared page caches
AIO page servers are designed to spread out the load of scanning the shared
page cache by interrupting the scan after a certain number of frames. This
helps to prevent a heavy load on the system during checkpoints and gives
the AIO page servers a chance to respond to any in-coming commands from
the stone.
The number of frames to scan before interruption is supposed to be scaled
so that a greater number of frames are scanned for larger shared page caches.
But due to this bug, the number of frames to scan is always set to the
default value of 1200. This results in a scan rate of about 1 second for
every 10MB of shared page cache space. For very large caches this can
result in checkpoints taking several minutes to complete, causing problems
when doing tasks such as backups which cannot be started during a checkpoint.
Workaround
If you have multiple extents, increase the number of AIO page servers running
via the stone configuration parameter STN_NUM_LOCAL_AIO_SERVERS.
blah blah
Last updated: 7/29/10