Slow checkpoints if much fewer AIO page servers than Free Frame page servers
If you increase the number of AIO page servers, but leave the number of
free frame page servers at 1, under some conditions of heavy load the checkpoints
may take a very long time to complete, an hour or more.
The problem is that the page servers operate on separate regions within
the cache, dividing the cache amoung the total number of page servers.
If there is only one free frame page server, it operates on the whole
cache, so it will return pages to one region at a time as it progresses
through the cache, and these free pages may be used for new work. On checkpoint,
when the AIO page servers write out all modified pages, most of these pages
will be in one region, so the remaining page servers will sit idle waiting
for one page server to complete writing.
Ensure that the SHR_NUM_FREE_FRAME_SERVERS is the same as STN_NUM_LOCAL_AIO_SERVERS.
In later versions this is the default.