Re: [p2pi] Real life torrent statistics

[Marshall Eubanks <[e-mail]>]

On Aug 19, 2008, at 2:48 AM, Stas Khirman wrote:

> > Here is a plot of the top 20 with a power law superimposed:
> >
> > http://www.americafree.tv/papers/torrent.top_20.png
> >
> > Regards
> > Marshall
> >
> [Stas Khirman]
>
> Marshall,
>
> Thank for your investigation, but unfortunately, when considering  
> tail of the distribution I'm not sure it power distribution can be  
> claimed - please check
> http://www.khirman.com/system/files/image/blog/torrent_distribution_blog_2.gif

Well, you generally can't tell anything from a linear-linear plots in  
these cases.

However, it is common in these cases for there to be a flattening with  
small rank
(thus the Pareto-Mandelbrot or Zipf-Mandelbrot distribution), and a  
lack of power as the numbers per bin get to
be of order unity. Both of these can be seen with the full data set

http://www.americafree.tv/papers/asn_groups.png
http://www.americafree.tv/papers/asn_groups.2.png

With only 3 degrees of freedom, the modified power law fits well for  
about 2 decades of rank.

This is not too surprising to me - similar Zipf type modified power  
laws can be seen in the selection
of video content (see http://www.imconf.net/imc-2007/papers/ 
imc78.pdf ) and I would
expect node selection to trace that.

Regards
Marshall

>>
> By popular demand, I just published major points and raw data on my  
> blog :
> http://www.khirman.com/blog/p2p_localisation
>
> Working notes available at http://www.khirman.com/system/files/image/blog/torrent_distribution.pdf
>
> Raw AS data http://www.khirman.com/system/files/image/blog/asn_groups.csv 
>  - please feel free to “twist” it – will appreciate any input.
>
>
> Regards
> Stas Khirman
> P.S. For those who in Silicon Valley – would be glad if you join our  
> Club event August 21 - http://www.khirman.com/ctc/20080821. Hot  
> controversial discussion and cold beer – guaranteed! (please sign-up  
> if you plan to attend)
>
>

_______________________________________________
p2pi mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/p2pi