April 8-16, 2008

Modeling P2P traffic

The AP389 is a p2p dominant AP. From this AP ploumidis isolated p2p sessions initialy from different applications and afterwards from applications that use the gnutella protocol. Ploumidis gave me 2 traces of downloaded P2P traffic

· in-session flow interarrivals

· flow sizes

The aim is to model the in-session flow interarrivals and the flow sizes. Do they follow a specific distribution as the traffic in web-dominant APs?

We wanted also to model the number of in-session flows – how many flows a sessions produces – but this was impossible (only 3 sessions were produced)

For more information http://www.csd.uoc.gr/~ploumid/ics/April%202008.htm

All Applications

In-session flow interarrivals

From the flowInterarrivalsP2Phomogenous.dat.gz trace I created the ccdf and QQplots for the distributions:

* Weibull

* Lognormal

* Exponential

* Extreme value

* Generalized extreme value

* Pareto

* Gamma

* Bipareto

Tsopanidis improved the initial matlab code I have created for the qqplots in order to run faster and produced the bipareto plots.

As we can see from the ccdf plot and the qqplot of the bipareto distribution with the emprirical data, in-session flow interarrivals for various p2p applications follow a bipareto distribution.

Flowsizes

From the _p2pflowsizes_homogenousSession389_thes90.dat.gz trace I created the ccdf and QQplots for the distributions:

* Weibull

* Lognormal

* Exponential

* Extreme value

* Generalized extreme value

* Pareto

* Gamma

* Bipareto

Macintosh HD:Users:itsomp:matlab:figures:linear:biparetolegend2.png

Macintosh HD:Users:itsomp:matlab:figures:linear:Exponential.png

As we can see no distribution fits our empirical data.

Flowsizes >1000 bytes

We also made the ccdf plot and qqplots for flow sizes>1000bytes:

April 17, 2008

Gnutella protocol

Ploumidis produced a trace with gnutella flows only. From this trace we will try to model p2p traffic for the flowsizes only since we have reached a model for in-session flow interarrivals.