Kbase P44280: BI takes a long time to format cluster but 0 buffers flushed
| Autor |
Progress Software Corporation - Progress |
| Acesso |
Público |
| Publicação |
05/11/2003 |
|
Status: Unverified
FACT(s) (Environment):
HP-UX 11.0
FACT(s) (Environment):
HP-UX
SYMPTOM(s):
bi takes 10 seconds to format cluster
0 buffers flushed at checkpoint
CAUSE:
OS Buffer Cache was taking a lengthy time to write down to disk which was delaying Progress from continuing.
FIX:
When a checkpoint occurs one of the actions performed by Progress is to issue a sync call to the OS to make sure the OS Buffer Cache is flushed (to make sure our data is written down to disk).
Review configuration of the syncer program which controls the frequency on HP that the OS buffer cache is sync'd.
By default syncer runs with a frequency of 30 seconds.
This may not be optimum if the disks on the system can not handle the amount of data being dumped from the OS Buffer Cache.
By increasing the frequency of the disk sync, there would be less to dump.
An alternative would be to decrease the BI Cluster size. Although this is contrary to normal practice, by decreasing the BI Cluster size the cluster would fill faster and trigger the checkpoint which would trigger the OS sync.