Comparing FCP and ECKD disk devices
In this final section we compare our results with FCP and ECKD devices to see if there are criteria which recommend one or the other disk connectivity type, based on the performance aspect.
Comparing normalized transactional throughput
The first criteria is the transactional database throughput reached. For this test we used the setup from each connectivity type which performs the best:
- For ECKD the HyperPAV setup with 20 aliases
- For FCP the multipath setup with policy multibus and
rr_min_io =100
The result is shown in Figure 1.
Observation
With 4 CPUs the throughput using FCP or ECKD disks is not very different. With 6 CPUs the throughput with FCP devices is increased by 50%. We did not do this test with the ECKD disk, because the setup with 4 CPUs had a utilization of only 80%.
Conclusion
With FCP devices higher rates of throughput can be achieved, but this requires more CPU capacity.
Comparing CPU cost per transactional throughput
The other interesting criterion is CPU cost. Figure 2 shows the CPU required to drive 1000 transactions.
Observation
With ECKD the lowest amount of CPU is required to drive a certain amount of workload. Using FCP requires around 14% more CPU. Be aware that the Y-axis starts at 90%.
Conclusion
ECKD devices required less CPU to drive the same amount of workload than FCP. There are two reasons for that:
- ECKD devices take advantage of the usage of SAP processors (which are not accounted for)
- For ECKD disks switching between the paths is part of the DASD driver and can be done very efficiently. For FCP devices a user space tool, the multipath daemon, is required which has a much longer path length.
As final criterion the administration effort should be considered. The handling of DASD and HyperPAV is much easier and less error-prone than the complex setup of FCP devices itself and the corresponding multipath setup.