On 10/13/2016 01:54 AM, Menaka Mohan wrote:
Hello Shyam,

Thanks a lot for such a detailed reply. I am new to this and was trying
hard to learn and got your help in the right time. After getting your
reply, I have experimented more and following are the findings.

I am clear with all the pre-requisites except only one. I searched for
any -easy to understand- examples online, but couldn’t find any. My
clients.ioz file has the “hostname  directory iozone_pathname”. But
where to specify the number of threads per client ? In the recent
commit, Number of threads are set to 4 in the script. But how to do that
while running our own iozone command ?

So, the number of lines in the clients.ioz should equal to, <threads> * <clients>. So say 'CLIENTS=hosta\ hostb', and threads are 4 (as in the script), then clients.ioz would have,

hosta <gluster-mount-point> <iozone-binary-path>
hosta <gluster-mount-point> <iozone-binary-path>
hosta <gluster-mount-point> <iozone-binary-path>
hosta <gluster-mount-point> <iozone-binary-path>
hostb <gluster-mount-point> <iozone-binary-path>
hostb <gluster-mount-point> <iozone-binary-path>
hostb <gluster-mount-point> <iozone-binary-path>
hostb <gluster-mount-point> <iozone-binary-path>

This should get things in order for you I believe.

So far, I am unable to find the cause of the following errors.

1)      extract_iozone_resuslt function returning None

2)      Line 13 in sync-drop-caches.sh “ source=${BASH_SOURCE[0]}”
returns Bad substitution.

But I am able to run the iozone commands in the script on my own and get
the results.

1)      # IOZone seq writes

2)      # IOZone seq reads

3)      # IOZone rand read / write

These look fine, except you are running with '-t 1', so if you correct the clients.ioz, then you should be able to run these with '-t 4', which would be the first thing to test.

Since we are getting the expected results, I assume covering all the
pre-requisites rightly except the above mentioned one.

While I run the actual gbench script, I get the following log. [
https://paste.fedoraproject.org/449225/63310741/ ]

The above, if you notice, is not getting any result from iozone, which is why the parser is failing to find 'Children see throughput' in the output, and you end up with the default result of "None".

I think if the clients.ioz is corrected, this should not occur. As stated earlier, let's get that fixed and see what results.

FWIW, the clients.ioz requirement to have as many lines per client is mentioned here [1], but possibly not very clearly (maybe another patch to the documentation :) ).

I also ran with iozone version 429 and a single line in my clients.ioz, the bench script failed, although the iozone output was a little more than what you see in your run. Let's hope that is some artifact of the setup and see if current suggestions work.

As I have checked, I also have bash shell as the default shell. [
https://paste.fedoraproject.org/449224/30577147/ ]

I am also trying to find the cause of the errors to my knowledge and
will report any progress soon. I have made few syntactic corrections in
the code and also updated the readme as you said in the previous email
and generated a pull request. Kindly review it.

Done, comments posted in github pull request.

Now, Please let me know what all you expect in the application other
than my understanding and the timeline and few initial contributions.

I would expect some results (assuming we are able to get this running and in order) and some analysis of any system resource across the setup (say memory or CPU or disk IO etc.), where you would need to investigate tool(s) to use and provide the results.

Note: I looked into few articles to reply properly into a thread in a
mailing list. I changed the subject as Re: Re: Sub. Still, If I am
unable to place it rightly in the thread, I am sorry. I will look into it.


Menaka M

[1] clients.ioz configuration, see iozone section in this page, https://gluster.readthedocs.io/en/latest/Administrator%20Guide/Performance%20Testing/

Or, search for "ioz.cfg contains these records" in the page above.
Gluster-devel mailing list

Reply via email to