Incorrect throughput being seen in report and differing PGL configs

[Dmitry Kurochkin]

Hi Mohammed.

On Thu, 04 Aug 2011 15:22:42 +0100, Mohammed Rakhada <morakhad at cisco.com> wrote:
> Hello Alex,
> 
> Thanks for your report, there isn't a problem here. It was my
> interpretation of the data. I didn't realise the throughput was averaged
> across the length of the test and so in my test as there was a slow ramp
> period (and it was a short test) it was giving me the "middle" value.
> All looks fine and as to be expected.
> 

Good to hear.

> Thanks to Dmitry for his initial input, meant to reply to him earlier
> but was just verifying some of my tests and results.
> 
> What would be nice would be to have a maximum and 95% measurement so
> that we can see what the actual measured peak was.
> 

May not be exactly what you want, but take a look at "load trace" plot
on "traffic rates, counts, and volumes" page.  Also, "everything" page
has stats for individual objects, e.g. "Object 'hits and misses'".

If you are interested in "raw" stats, you can get it with
polygraph-lx(1) and polygraph-ltrace(1) tools.  E.g.:

  $ ltrace --win_len 30sec --side clt --objects rep.rate LOG

Would give you reply rate stats with 30sec interval.  Polygraph stat
cycle length is 5sec by default and can be changed with --stats_cycle
option.

Regards,
  Dmitry

> As for the hosts and differing configs I've implemented the changes
> Dmitry suggested and they work fine.
> 
> Thanks
> 
> Mohammed Rakhada
> 
> On Thu, 2011-08-04 at 08:12 -0600, Alex Rousskov wrote:
> 
> > On 08/02/2011 11:15 AM, Dmitry Kurochkin wrote:
> > 
> > > On Tue, 02 Aug 2011 14:51:09 +0100, Mohammed Rakhada <morakhad at cisco.com> wrote:
> > 
> > >> throughput: 4196.00xact/sec or 608.43Mbits/sec
> > >>
> > >> however when I am looking at the switch statistics the number reported
> > >> is much lower (360Mbits/sec)
> > >>
> > >> Could you clarify what the throughput value actually relates to?
> > >>
> > > 
> > > Throughput on the index page of HTML report is client side reply
> > > throughput, i.e. (size of all replies clients received) / duration).  It
> > > does not include requests or replies sent by servers.
> > > 
> > > I am not sure why you see lower throughput stats on the switch.  You may
> > > get wrong stats in reporter if you specify a single log multiple times
> > > on the command line.  I do not think it is likely, but this may be a bug
> > > in Polygraph reporter or client.
> > 
> > Another possibility here is that the switch is counting traffic volumes
> > over a longer (or shorter!) periods of time while Polygraph reporter is
> > using response volume during the specified test phase(s).
> > 
> > If you want to investigate this further, running longer tests with fixed
> > response sizes while looking at runtime switch stats may be useful.
> > 
> > HTH.
> > 
> > Alex.
> 
> 
> _______________________________________________
> Users mailing list
> Users at web-polygraph.org
> http://www.web-polygraph.org/mailman/listinfo/users