Ceph performance measurement task

[Denis Klepikov]

I would like to have rally test for ceph performance based on numbers of instances with hi-loaded disk operations.
This test can allow to determine how much instances we can run on tested environment (In the context of the choice of the lowest acceptable disk performance (KiB/s, IOPS) on instance).

How to get statistics manually:
1 Create cluster with ceph
2 Create instance i(1) into cluster
3 on instance run command

rm -rf /tmp/test && spew -i 50000 -v -d --write -r -b 4096 10M /tmp/test
-i 50000 may be less if iterations
You will get an output like
Iteration: 1 Total runtime: 00:00:07
WTR: 1413.41 KiB/s Transfer time: 00:00:07 IOPS: 353.35

Iteration: 2 Total runtime: 00:00:14
WTR: 1484.27 KiB/s Transfer time: 00:00:06 IOPS: 371.07

Iteration: 3 Total runtime: 00:00:21
WTR: 1493.15 KiB/s Transfer time: 00:00:06 IOPS: 373.29

Iteration: 4 Total runtime: 00:00:28
WTR: 1461.88 KiB/s Transfer time: 00:00:07 IOPS: 365.47

Iteration: 5 Total runtime: 00:00:34
WTR: 1465.91 KiB/s Transfer time: 00:00:06 IOPS: 366.48

Save it with time stamp on any iteration

date, time, instance#, WTR, Transfer time, IOPS, total instances#

4 Create instance i(2) into cluster
run the same as in p.3
5 Create instance i(n) into cluster
run the same as in p.3

Create the output info based on data

Into output would like to see graphics :
1 Main ceph performance with # of started nodes in KiB/s or MB/s
Y-axis WTR (WTR=(WTR(i1)+WTR(i2)+...+WTR(i(n)))
X-axis # of started nodes with spew running

1.1 Main ceph performance with # of started nodes in IOPS on Y-axis

1.2 Main ceph performance with # of started nodes Transfer time on Y-axis

1.3 Average ceph performance per instance
Y-axis WTR (WTR=(WTR(i1)+WTR(i2)+...+WTR(i(n)))/(# of started instances)
X-axis # of started nodes with spew running

1.4 Main ceph performance per instance in IOPS on Y-axis

1.5 Main ceph performance per instance Transfer time on Y-axis

2 Node # ceph performance
Y-axis WTR
X-axis # of started nodes with spew running

2.2 Node # ceph performance with # of started nodes in IOPS on Y-axis

2.3 Node # ceph performance with # of started nodes Transfer time on Y-axis

In case of additional clarification please contact me via e-mail or skype.

[Boris Pavlovic (boris-42)]

This can be already done using VMTask.boot_run_command

https://github.com/stackforge/rally/tree/master/doc/samples/tasks/scenarios/vm

You just need to pass you .sh script that will run proper command like this:
https://github.com/stackforge/rally/blob/master/doc/samples/tasks/support/instance_dd_test.sh

So only thing that you need is to make such file and run benchmark.