|
Understanding HyperPAV with IntelliMagic Vision
As more of customers move to HyperPAV I would like to share some experience with HyperPAV management. HyperPAV does not have many controls; it is mostly doing its work silently in the background. If you drive your storage hard enough, however, queues may build up.
The most important metric is the HyperPAV Wait Rate; it shows how many I/O requests could not be served immediately because all available aliases were in use. If the wait rate becomes high (hundreds, more than a few percent) you may start to see IOSQ time develop. At that point, it becomes attractive to assign more aliases to the LCUs that show problems.
It may seem counter intuitive, but you can also reduce the HyperPAV Wait Rate by migrating to larger logical volumes: if you move from 192 base 3390-27 with 64 aliases to 96 base 3390-54 with 160 aliases the HyperPAV Wait Rate will become less because z/OS has more freedom
to schedule I/Os.
Another metric that is available at the LCU level is the maximum HyperPAV Queue Length; this is of less value, however, since even very short spikes may cause this number to increase. Also, even if
you want to investigate what causes the spike, there is no equivalent metric available at the device level, so you will never to be able to correlate the spike to a workload problem. Only when IOSQ time starts to develop will you need to investigate more.
It should be noted that not all HyperPAV contention can be fixed; when there are multiple updates to the same extent (data set), they must be serialized.
With IntelliMagic Vision you can easily monitor HyperPAV activity. The product will also make recommendations, such as how many aliases will be needed for your workload. Some samples are provided in our technical note on this topic. download to read full article
|
