High Kernel usgae for sysbase server

Hi alpmoon,

Only two sybase processes are very active and the other sybase  processes consumes 2% to 3% CPU. The value of the context switch is not large when the users report slow response.

I will move the most sybase process to a new AIX server but in the meantime I have to resolve the slow response problem. Do you have interium solution to resolve the performance issue.  Right now we re-schedule all non-ciritical batch processing of another sybase to non-peak hours and the performance is improved. However we cannot stop the users of other sybase to do on-line processing (ie enquiry or update

<<<
CPU usage can be restricted based on application etc
>>>
How can I restrict a sybase application to use CPU?

cpc

I think the most active Sybase instance (dataserver) needs to use more than 1 CPU. First you should set "max number of engines" parameter to 2. Depending on the version of ASE, most probably you need to set "online engines at startup" to 2 as well. I can tell more detail if you provide ASE version.

However, I am not sure whether this configuration change affect other instances or not. That was the reason I suggested to have one or at most two Sybase instances on a two CPU UNIX box. If others are not really active, it should be OK. But the second active ASE instance may be affected by this change.

As I said, the better solution is moving all databases to one ASE instance if it is possible. In this case Sybase control all processes and it is much more efficient than leaving control to UNIX. Handling context switches through UNIX is much more expensive than doing it through Sybase.

After setting two engines you can restrict some users to use only one engine at a time. It is only meaningful if you move all databases into one ASE dataserver.

Users reports slow response at Sybase A on 7th March and stock price update was delayed almost by 10 minutes. At that moment Syabse B was also very active.  Compare the CPU utllization on 6th March and that day no user report slow response at Sybase A and the CPU utilization of Sybase B was very low.

Sybase A
                                         CPU Time                
06/03/07 11:02:10 - 10:02:10  25:33 (50%)
06/03/07 10:02:10 - 10:32:10  25:38 (50%)

Sybase B
06/03/07 11:02:10 - 10:02:10    7:02 (14%)
06/03/07 10:02:10 - 10:32:10  19:39  (38%)
 
Sybase A
                                         CPU Time                
06/03/06 11:02:10 - 10:02:10  25:03 (50%)
06/03/06 10:02:10 - 10:32:10  26:07 (50.5%)

Sybase B
06/03/06 11:02:10 - 10:02:10   0:56 (1%)
06/03/06 10:02:10 - 10:32:10   1:24 (2.5%)

It seems that the slow response of Sybase A was related to Sybase B was very active. However it does not make sense, as the CPU utilization of Sybase A is almost the same. If the CPU utilization of Sybase A drops when Sybase B is very active, I will be convinced that it is CPU capacity. The other two sybase (ie Sybase C and Sybase D) only used less than 3% of the CPU. It seems that the bottle neck is within sybase A update process (ie it is stock price update, the stock price is update three times per minutes and it was backlog for 10 minutes) . It maybe related to I/O or database locking.
The system has almost no paging and the problem is not related to memory.

How can I find out the I/O update performance bottle neck of Sybase A? I am not convice myself it is related to CPU. Is it possible that the value of network packet size is too low?

I haven't resolved the problem and I want to keep this question for one more month. Maybe some experts have new suggestion to my problem.