DKB
Member
Hi,
We are a brand new EnterpriseOne site running version 8.10 on a 5-way 870 with 16GB memory and 22 disk arms. The WebSphere environment is running on Intel.
We have recently run some stress tests (before we go live) which has shown the memory and disk setup on the iSeries to be way short of what's required. I am addressing this memory and disk shortage. We will ultimately be running several hundred HTML users and I would like to get people's thoughts/guidance on the jde.ini settings and other assumptions we've made (after reading the relevant IBM redbook)...
- 1 call object kernel for every 7.5 users
- 1 security kernel for every 100 users
- 1 network kernel for every 20 call object kernels
- checkKernelHealth set to 5000
- useAS400Security set to FALSE
- Each HTML will generate upto 5 JDBC server jobs (QZDASOINIT)
- Each ODBC job will need 15MB memory
- Each UBE will require upto 400MB memory
For information, the stress tests we ran showed that we were getting millions of 'Teraspace EOA exceptions' in every performance interval (over 7.5million in a 1 minute sample!) which was consuming huge amounts of CPU. These can be seen by viewing the Component Report within Performance Tools. Applying PTF 5722SS1/SI17614 solved this problem and reduced the exceptions down to 9,000...CPU consumed reduced by approx. 50%! We also found that separating the QZDASOINIT jobs out of *BASE memory pool greatly improved things, although we are still significantly short of memory and disk.
Any guidance/thoughts on our assumptions would be greatly appreciated, particularly with regards to exactly how many kernels we should be running and recommendations for the numbers of kernels/jobs per available CPU.
Thanks,
Dean.
We are a brand new EnterpriseOne site running version 8.10 on a 5-way 870 with 16GB memory and 22 disk arms. The WebSphere environment is running on Intel.
We have recently run some stress tests (before we go live) which has shown the memory and disk setup on the iSeries to be way short of what's required. I am addressing this memory and disk shortage. We will ultimately be running several hundred HTML users and I would like to get people's thoughts/guidance on the jde.ini settings and other assumptions we've made (after reading the relevant IBM redbook)...
- 1 call object kernel for every 7.5 users
- 1 security kernel for every 100 users
- 1 network kernel for every 20 call object kernels
- checkKernelHealth set to 5000
- useAS400Security set to FALSE
- Each HTML will generate upto 5 JDBC server jobs (QZDASOINIT)
- Each ODBC job will need 15MB memory
- Each UBE will require upto 400MB memory
For information, the stress tests we ran showed that we were getting millions of 'Teraspace EOA exceptions' in every performance interval (over 7.5million in a 1 minute sample!) which was consuming huge amounts of CPU. These can be seen by viewing the Component Report within Performance Tools. Applying PTF 5722SS1/SI17614 solved this problem and reduced the exceptions down to 9,000...CPU consumed reduced by approx. 50%! We also found that separating the QZDASOINIT jobs out of *BASE memory pool greatly improved things, although we are still significantly short of memory and disk.
Any guidance/thoughts on our assumptions would be greatly appreciated, particularly with regards to exactly how many kernels we should be running and recommendations for the numbers of kernels/jobs per available CPU.
Thanks,
Dean.