Originally posted by: Pauld D. Hunter
We have a z/9 BC system running z/OS 1.7. We run two major applications, a hospital billing system and a payroll system. Performance has always been excellent. We have 4 2Gbps FICON channels to the 9980 and service times are usually from .5 ms up to 4 ms with the higher the IOPS the better the response time. All of a sudden over the last few weeks we hit some times where the service times on some volumes climb to 70 ->80 _> 90 ms with some even showing into the hundreds of ms. Needless to say, this is catastrophic for our users.
We have uploaded logs from the SVP to HDS and they analyzed them and said we are experiencing periods where the cache write pending gos over 30%, sometimes 40% and even 50%. They also pointed to some LDEVs that were very busy and said if we moved them that could resolve the problem or we could add cache. We have 36GB of cache and there are 28TB on the box with about 3 TB of mainframe DASD, mostly 3390-3 emulation. We were thinking of going up to 48GB of cache. Adding cache is the quickest, least disruptive but most expensive and the LDEVs mentioned aren't on the mainframe (in fact there were no hot spots or over utilized array groups on the mainframe allocations). Has anyone out there experienced anything similar? We have been on this box since 2003 and have never had a DASD performance issue until 10 days ago. Any input would be appreciated.