The use of PAV has effects on both the users’ perception and on the system and hardware.
Effects from the user viewpoint
PAV primarily reduces the IO times (software I/O time):
In the case of synchronous IOs, copy processes, batch mode, etc., the throughput then also improves correspondingly. The users perceive shorter runtimes (if the IOs had been constricted before due to parallel load).
In OLTP mode, the throughput is specified by the user-generated work and will at most increase to a small degree. The users perceive shorter transaction times.
Basically the user perception depends on how high the proportion of the IO wait times is in the total runtime and transaction time.
Effects on the system and hardware
With PAV, more IOs can be processed in the same time, thus condensing certain loads. This can have an effect on the utilization of the CPU, disk peripherals, and channels.
CPU
The shortened IO wait times enable more data processing to take place in the same time; the CPU utilization can thus increase. This is merely a consequence of the compressed load. The increased CPU utilization by PAV itself is negligible; the CPU utilization as a whole is as a rule not influenced.
Disk peripherals
The requirements with respect to the storage system and the RAID group increase, and in consequence the hardware IO times also increase. Slight increases are an unavoidable consequence of PAV and constitute the expected behavior as long as the increase in the hardware I/O time is significantly lower than the reduction in the software I/O time, see also the measurements in section "Performance when PAV is used".
However, if the corresponding RAID group does not have sufficient reserves, a new bottleneck can occur here (above all when the cache hit rate is low), which means that the performance enhancement can be lower than expected. This can be recognized from the sharply increasing hardware IO times. In such a case, the following measures can make sense in order to increase the volume’s raw performance:
distribution of the data to one or more RAID groups with a lower workload,
reconfiguration of the RAID group to a RAID level with a higher performance,
striping over more disks, or
in extreme cases switching to faster hard disks or a larger cache size.
In all cases the reason for the bottleneck should be understood beforehand. The first indications for this can be provided, for instance, by the cache hit rate, the IO rate, the IO size, and the ratio of write to read IOs. It should above all be clarified whether the RAID group is fully utilized by a particular load on one volume, or if the overload is a result of multiple loads (possibly also of other systems) on various volumes in the RAID group.
Channels
The channels are also utilized better by PAV. After PAV activation, the increase in the channel utilization should also be checked. Possibly further paths should be made available, above all in the case of throughput-oriented loads.