- The real physical I/O performance is worse on almost every system seen so far
- z/OS volume queuing is often a problem when everything is defined as a 3390-3, or larger
- Having gigabytes of cache often bails you out, but you still do an I/O… pool tuning is critical
- Heavy physical write penalty
– Overload the NVS cache and the world stops
There is no silver bullet for DASD performance, and the manufacturers claims are all greatly overstated. Based on observations at many installations, the sole saving factor of these systems is the huge cache on the controller. The actual device performance for a Synch read of one 4K page from disk is generally worse than a native 3390, and write performance is very poor compared to a 3390. Now putting the write side into perspective, most writes take place into an NVS cache – so the delay seen by the application is often low to none. However, when really high volume write situations occur, the cache my fill to a critical internal threshold because the true physical writes take a long time. When this happens, the device enters an ‘emergency de-stage’ scenario where writes have priority – and all read access times can double or triple.