object size

[Dmitry Kurochkin]

shahab bakhtiyari <shahab371 at gmail.com> writes:

> Thank you Dimitry
>
> well I dont think that I'll go for making my own distribution,  rather I
> use the existing ones. But I need a little documentation about existing
> ones,  like   zipf(64)  , I have no idea how large  it is? the  only thing
> I found says :  "Zipf(1): *zipf(world_size)*" .  its alittle bit  unclear
> for me,  is there any documentation for that?
>

A list of distributions and their parameters is available at [1].  You
can find information on a particular distributions on wikipedia or other
resources (e.g. Zipf's law [2]).  Note that Polygraph's implementation
may not match exactly the mathematical formula.  You can test particular
parameters for Polygraph distributions using distr-test tool (installed
as polygraph-distr-test(1)), e.g. to see what values zipf(64) would
produce run:

  $ polygraph-distr-test --distr 'zipf(64)'

It would generate 100000 values (by default) and print histogram, mean,
min, max values and some other information.

Regards,
  Dmitry

[1] http://www.web-polygraph.org/docs/reference/pgl/types.html#type:docs/reference/pgl/types/distr
[2] http://en.wikipedia.org/wiki/Zipf's_law

> On 11 May 2012 18:39, Dmitry Kurochkin <
> dmitry.kurochkin at measurement-factory.com> wrote:
>
>> Hi Shahab.
>>
>> shahab bakhtiyari <shahab371 at gmail.com> writes:
>>
>> > Hi guys
>> >
>> > Yet another question for me(thank you very very much for previous
>> > responses),
>> >
>> > I am wondering if the objects' size used in webaxe-1(or generally in all
>> )
>> > workload(s)  are realistic? or how much they are close to true size?
>> >
>> > I mean ,  using exp(4.5kb) with mean4.5kb and  max 53kb   for image
>> >  objects  sounds too little, is'nt that?
>> > or 300kb for downloads? am I totally wrong?
>> >
>>
>> You may be right.  I do not think these sizes match average object or
>> image on the Internet.  But usually what you are interested in is
>> simulating *your* traffic properties.
>>
>> The numbers in the provided workloads should be a good starting point.
>> But for best results you should create distributions that match your
>> needs.  See [1] for details on user-defined distributions.  Creating
>> proper distributions for your tests may be difficult.  You will need
>> some existing data for it (e.g. Squid access logs).  Then you analyze it
>> with some auxiliary tools to get percentages for distributions (for
>> Squid access logs you might use access2pgl tool from
>> src/tools/access2poly/).
>>
>> Regards,
>>  Dmitry
>>
>> [1] http://www.web-polygraph.org/docs/reference/tabdistr.html
>>
>> > thank you
>> > --Shahab
>> > _______________________________________________
>> > Users mailing list
>> > Users at web-polygraph.org
>> > http://www.web-polygraph.org/mailman/listinfo/users
>>