Question & Answer
tsmdiskperf.pl is a tool that is provided as part of the IBM Spectrum Protect Blueprint configuration scripts (see related link section). What is the blocksize when generating load using the script?
The tsmdiskperf tool bases it measurements from data that is reported by iostat, so there is no assumption about the block size in the measurements.
The average IO size can be determined dividing the average combined throughput by the average IOPS.
For example, here is the result of a database workload:
: Devices reported on from output:
: Average R Throughput (KB/sec): 168358.03
: Average W Throughput (KB/sec): 150517.40
: Avg Combined Throughput (MB/sec): 311.40
: Max Combined Throughput (MB/sec): 619.09
: Average IOPS: 39637.21
: Peak IOPS: 79219.57 at 04/18/2018 10:34:22
: Total elapsed time (seconds): 13
And we can calculate average IO size as:
311.40 * 1024 / 39637.21 = 8.04 KB
Below is a snapshot of iostat that was running while the above workload was in progress. In the iostat output, the avgrq-sz value should match what we just calculated. For this column, the unit is 512 byte sectors, so a value of 16 is equal to 8KB.
As the calculations are based on iostat, the numbers will reflect additional IO that is generate by an IBM Spectrum Protect server that is configured to use the same filesystems as the tsmdiskperf tool.
17 June 2018