We have an in-house tool that monitors CPU consumption and other performance metrics for address spaces and CICS transactions. This data is summarized every 5 minutes to allow easy querying and comparisons. All of this data is stored in DB2 and we have created some impressive SQL queries to quickly and efficiently display anomalies and critical issues. Our management runs our mainframe near full capacity to achieve "full value" for their processors. Needless to say, we don't always have enough tolerance to allow CPU consumer "outliers".
With this tool, researching high CPU consumers that are out of their average have become easy, but we cannot easily determine the perceived impact. Just because an address space is consuming more CPU and the system is at 100% doesn't necessarily mean our customers are perceiving any negative impact. The system could be over its CPU share for the sysplex and another system may have spare CPU to 'steal'. Having the ability to see any delays in our lower service class address spaces would be a great addition. I've been working on a major enhancement to collect CPU delays (I really don't care about other types of delays at the moment) for each address space, but I'm not getting the desired results. I read that IWMRQRY and IWMCOLL appear to be the best services for this solution. However, I found that neither of these services return an ASID or any identifier of an address space. IWMRQRY has an input parameter for ASID, but running this service for every active ASID on the system (roughly 150 on our sandbox system) every quarter second (our WLM interval) is not cheap. Is there a better solution for collecting this data or am I missing something about these WLM services? Thank you, Brian Chapman ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
