The following measurements were conducted with 1 volume via 1 channel path, synchronous IOs and 100% cache hit rate. They show the effect of FastDPAV on IO times and, thereby, IO rates and throughput.
Tasks | FastDPAV | random25 | |||||
IO/s | SW Time | HW Time | Mbyte/s | SW Time | HW Time | ||
3 | 0 | 5,785 | 0.3 | 0.1 | 299 | 4.4 | 1.5 |
6 | 13,675 | 0.1 | 0.1 | 672 | 1.8 | 1.8 | |
8 | 0 | 5,781 | 1.2 | 0.1 | 638 | 12.2 | 1.5 |
6 | 23,776 | 0.2 | 0.2 | 1,103 | 3.0 | 2.7 |
FastDPAV shows the same behaviour as static PAV. The parallelization of IO requests leads to significantly lower IO times an therefore to a higher IO rate and higher throughput. In most application cases FastDPAV can work witch less alias devices than a configuration with static PAV would need in total.
The increasing hardware IO times are not caused directly by FastDPAV, but are a result of the increased throughputs, and thus of the higher workload of the channel and storage system.