On 10/15/2016 09:14 PM, Menaka Mohan wrote:

I have been able to successfully run the IOZone bench test on the
Gluster setup. Kindly find the results here [1].

Neat, thanks.

But for the smallfile distributed I/O benchmark, the test fails. Since
we are taking 5 samples in each operation, the " --operation create "
fails reporting that the file already exits from the second
sample [1462.758118, None, None, None, None] . Necessary changes are to
be made such that the --operation create happens only once and others (
--operation read and --operation ls-l ) run as usual. Kindly correct me,
If I am wrong.

I am wondering why [3] is not getting triggered and cleaning up the files in your test run. It seems to be doing the job in my setup. Possibly something to check/debug when you get a chance.

Assuming that as the reason, I have made the necessary changes to the
code [sample _size for create operation is 1] and ran the bench test.
Kindly find the results here [2].

This should be fine for the exercise. The interest in running a test more than once is to ensure we do not have any noise from a single run that gives us incorrect numbers as a result. So for now, running this once is fine, unless you get some time to debug as above.

I will further work on it as a single script and post the output as
mentioned in the GitHub page. Since this is a very time consuming
process, I will simultaneously work on the next task which is to
investigate the tools to use and analyze the system resource usage
across the setup.

I would encourage looking at the next part (i.e tools to analyze the system) of this task, as that can provide some more meaningful data for the project.

[1]   https://paste.fedoraproject.org/451045/55349814/
[2]   https://paste.fedoraproject.org/451953/14765795/

[3] GlusterBench line that should cleanup small files: https://github.com/gluster/gbench/blob/master/bench-tests/bt-0000-0001/GlusterBench.py#L257
Gluster-devel mailing list

Reply via email to