Symptom
Work processes are restarted periodically with the parameter
rdisp/wp_auto_restart. If in addition to this, the parameter
rdisp/noptime is set to a value bigger than 0, many work processes can
be restarted at the same time.
The 7.20 Kernel brought new, CPU-intensive source code into the start-up phase of the work process, so that rdisp/noptime > 0 may cause a high CPU consumption when restarting work processes.
The 7.20 Kernel brought new, CPU-intensive source code into the start-up phase of the work process, so that rdisp/noptime > 0 may cause a high CPU consumption when restarting work processes.
Other Terms
CPU spikes rdisp/noptime rdisp/wp_auto_restart
Reason and Prerequisites
Reason
High CPU load when starting a work process and restarting many work processes at the same time.
Prerequisites:
Kernel 7.20 or higher.
High CPU load when starting a work process and restarting many work processes at the same time.
Prerequisites:
Kernel 7.20 or higher.
Solution
Workaround:
Since this problem occurs only due to the synchronization with rdisp/noptime, you should set the parameter to the value 0. The next step is to check if the CPU peaks were caused by restarting work processes.
Solution:
After you implement the correction provided by this SAP Note, rdisp/noptime > 0 restarts the work processes with at least three seconds delay respectively. For this, the new instance profile parameters
rdisp/nop_method = [0,1]
rdisp/nop_period = value in seconds
are introduced. With the combination
rdisp/noptime > 0
rdisp/nop_method = 1 you activate the new procedure. If rdisp/nop_method has the value 0, the old procedure is used.
You can use rdisp/nop_period = n (seconds) to vary the delay between the work processes (default = 3 seconds). Caution: Too long time intervals for rdisp/nop_period may cause the complete restart period (rough estimate = number of WPs times rdisp/nop_period) becomes bigger than rdisp/noptime (see SAP Note 158346, or better the RZ11 documentation).
NOTE:
The correction was delivered with Patch Level 720/414 or 721/43 It contains an error so that the dispatcher terminates immediately on the startup. Currently, this has been determined for the operating system Linux only. We provide a correction. As a workaround, set the parameters in the instance profile:
rdisp/nop_method = 0 or 1
rdisp/nop_period = 5
Since this problem occurs only due to the synchronization with rdisp/noptime, you should set the parameter to the value 0. The next step is to check if the CPU peaks were caused by restarting work processes.
Solution:
After you implement the correction provided by this SAP Note, rdisp/noptime > 0 restarts the work processes with at least three seconds delay respectively. For this, the new instance profile parameters
rdisp/nop_method = [0,1]
rdisp/nop_period = value in seconds
are introduced. With the combination
rdisp/noptime > 0
rdisp/nop_method = 1 you activate the new procedure. If rdisp/nop_method has the value 0, the old procedure is used.
You can use rdisp/nop_period = n (seconds) to vary the delay between the work processes (default = 3 seconds). Caution: Too long time intervals for rdisp/nop_period may cause the complete restart period (rough estimate = number of WPs times rdisp/nop_period) becomes bigger than rdisp/noptime (see SAP Note 158346, or better the RZ11 documentation).
NOTE:
The correction was delivered with Patch Level 720/414 or 721/43 It contains an error so that the dispatcher terminates immediately on the startup. Currently, this has been determined for the operating system Linux only. We provide a correction. As a workaround, set the parameters in the instance profile:
rdisp/nop_method = 0 or 1
rdisp/nop_period = 5
Header Data
| Released On | 26.02.2013 10:11:20 | ||
| Release Status | Released for Customer | ||
| Component | BC-CST-DP Dispatcher, Task Handler | ||
| Other Components |
| ||
| Priority | Recommendations / Additional Info | ||
| Category | Performance |
No comments:
Post a Comment