Summary information |
|
|---|---|
Default value |
200 |
Range of values |
1–2147483647 |
Status |
Dynamic |
Display level |
Comprehensive |
Required role |
System Administrator |
cpu accounting flush interval specifies the amount of time, in machine clock ticks, that Adaptive Server waits before flushing CPU usage statistics for each user from sysprocesses to syslogins, a procedure used in chargeback accounting. (Note that this is measured in machine clock ticks, not Adaptive Server clock ticks.)
When a user logs in to Adaptive Server, the server begins accumulating figures for CPU usage for that user process in sysprocesses. When a user logs off Adaptive Server, or when the value of cpu accounting flush interval is exceeded, the accumulated CPU usage statistics are flushed from sysprocesses to syslogins. These statistics continue accumulating in syslogins until you clear the totals using sp_clearstats. You can display the current totals from syslogins using sp_reportstats.
The value to which you set cpu accounting flush interval depends on the type of reporting you intend to do. If you intend to run reports on a monthly basis, set cpu accounting flush interval to a relatively high value. With infrequent reporting, it is less critical that the data in syslogins be updated as often.
On the other hand, if you intend to do periodic ad hoc selects on the totcpu column in syslogins to determine CPU usage by process, set cpu accounting flush interval to a lower value. Doing so increases the likelihood of the data in syslogins being up to date when you execute your selects.
Setting cpu accounting flush interval to a low value may cause processes to be mistakenly identified as potential deadlock victims by the lock manager. When the lock manager detects a deadlock, it checks the amount of CPU time accumulated by each competing processes. The process with the lesser amount is chosen as the deadlock victim and is terminated by the lock manager. Additionally, when cpu accounting flush interval is set to a low value, the task handlers that store CPU usage information for processes are initialized more frequently, thus making processes appear as if they have accumulated less CPU time than they actually have. Because of this, the lock manager may select a process as the deadlock victim when, in fact, that process has more accumulated CPU time than the competing process.
If you do not intend to report on CPU usage at all, set cpu accounting flush interval to its maximum value. This reduces the number of times syslogins is updated and reduces the number of times its pages need to be written to disk.
