[Haizea] Final CPU Utilization

[Mehdi Sheikhalishahi]

So in analyzing full raw data, we should do something like that you did
in haizea_swf2lwf class to remove ramp-up/ramp-down effects, Am I right?

Some more questions about that code in haizea_swf2lwf:

1. By introducing no_ramp_cutoff, we take care of the beginning 5% of
workload? and also the end of workload?

no_ramp_cutoff = from_time + ((to_time - from_time) * 0.05)

2. In utilization_no_ramp calculation, we shouldn't consider if there is no
waiting time for a job, then reduce interval. And why we are using this
technique? we want to put jobs in the normal condition of a workload that is
in the middle i.e. neither the beginning nor the end.

if wait_time != -1 and submit_time + run_time >= no_ramp_cutoff:
                    start_in_interval = max(no_ramp_cutoff, submit_time)
                    end_in_interval = min(to_time, submit_time + run_time)
                    time_in_interval = end_in_interval - start_in_interval
                    utilization_no_ramp += time_in_interval * num_processors

Is it possible for me to access R infrastructure now?

Regards,
Me

On Tue, Jun 15, 2010 at 8:52 PM, Borja Sotomayor <borja at borjanet.com> wrote:

> Hi,
>
> > Interesting issue with utilization.
> > So you mean because of ramp down effect even
> > CPUUtilizationProbe.COUNTER_UTILIZATION counter will not be a very good
> > parameter to show CPU Utilization in general?
>
> It's not a good parameter if you look at individual values (even the
> running average) because those values are going to suffer from a
> ramp-down effect (in fact, there's also a ramp-up effect: at the start
> of the experiment, there will be an initial period of underutilization
> until enough requests have been made). I do find it useful for
> debugging purposes: it's a quick way to check if utilization is what I
> expect it to be (specially if I'm testing an algorithm that's supposed
> to result in higher utilization)
>
> To get a more accurate utilization metric, it's better to rely on the
> full raw data (i.e., all the utilization values throughout the entire
> duration of the experiment) so you can process it and remove ramp-up
> and ramp-down effects. I typically do this with R (mostly because it's
> what I use to generate all my graphs), but it shouldn't be hard to do
> in Python.
>
> Cheers!
> --
> Borja Sotomayor
> PhD Candidate in Computer Science, University of Chicago
> http://people.cs.uchicago.edu/~borja/
> _______________________________________________
> Haizea mailing list
> Haizea at mailman.cs.uchicago.edu
> https://mailman.cs.uchicago.edu/mailman/listinfo/haizea
>



-- 
Regards,
Mehdi
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.cs.uchicago.edu/pipermail/haizea/attachments/20100616/c06f0982/attachment.htm