Poisson Pareto Burst Process

This is a proposed new ns-3 traffic generator.  The PPBP is a process based on the overlapping of multiple bursts with heavy-tailed distributed lengths.  Events in this process represent points of time at which one of an infinite population of users begins or stops transmitting a traffic burst.  The PPBP is closely related to the M/G/∞ queue model.

[Tommaso Pecorella, 2011-10-31 14:16:39]

Hi,

to be honest I have several concerns with this.

First and foremost, the code is unclear. PPBP is per-se a multi-state system,
where you have a "master" state launching new "generators" and sub-states (one
for each generator). This is not clear in the code.
This of course might lead to an unknown number of sub-generators, potentially a
high number. In a generic code you should try to void this by setting a maximum,
even if this might screw the "right" stats. It should be up to the user to
accept the risk of a memory explosion or to have sub-optimal stats.
I'd put a "max-sub-processes" number and, if 0 there's no limit, if it's
positive, there is.

Moreover, I fear that the code doesn't really mimic the PPBP, check my comment
in the code.

Typographically, you have used non-ascii characters in the source. remove them
all. Plus there are tabs here and there. Also to be removed.

Last but not least, for such a piece of code, it's totally needed a regression
test, in order to validate numerically that the output is consistent with the
assumptions. As is, to show that the app is effectively generating a long-tail
distributed output.

By the way, you might want to take a look at:
http://codereview.appspot.com/224079/
it's an old and never included set of traffic generators. It's nothing like
PPBP, but you might find useful ideas.

Cheers,

T.

http://codereview.appspot.com/4997043/diff/1/src/applications/doc/ppbp.rst
File src/applications/doc/ppbp.rst (right):

http://codereview.appspot.com/4997043/diff/1/src/applications/doc/ppbp.rst#ne...
src/applications/doc/ppbp.rst:22: λp:	mean rate of burst arrivals
Avoid to use non-ascii characters. Either use "lambda" or (better) use the LaTeX
syntax in the documentation. The second option is way better for Doxygen.

Not commenting other occurrences, but it's spread among all the code and not
only for lambda (e.g., also infinity symbol)

http://codereview.appspot.com/4997043/diff/1/src/applications/doc/ppbp.rst#ne...
src/applications/doc/ppbp.rst:27: E[n] = Ton x λp
It's a formula, use LaTeX syntax.

http://codereview.appspot.com/4997043/diff/1/src/applications/model/ppbp-appl...
File src/applications/model/ppbp-application.cc (right):

http://codereview.appspot.com/4997043/diff/1/src/applications/model/ppbp-appl...
src/applications/model/ppbp-application.cc:161: if (m_offPeriod)
I didn't check the code enough to be sure about the consistency of this, but
from my experience it's dangerous. Use an uint to store the state and check if
it's even or odd. In this way you'll avoid overlapping states issues (with mixed
pareto / poisson state length it is possible).

Moreover, what happens if it goes off (as is, false)? Unclear.

http://codereview.appspot.com/4997043/diff/1/src/applications/model/ppbp-appl...
src/applications/model/ppbp-application.cc:232: double data_rate = (double)
nextTime.GetSeconds () / m_activebursts;
Wrong. In PPBP the data rate of single sources is constant. What contributes to
the *mean* data rate is the sub-sources generation rate.

http://codereview.appspot.com/4997043/diff/1/src/applications/model/ppbp-appl...
src/applications/model/ppbp-application.cc:251:
PPBPApplication::ScheduleStopEvent ()
What's the meaning of an empty function ?

[Tommaso Pecorella, 2011-10-31 14:18:29]

Sorry, forgot to add an useful link.

http://www.scribd.com/doc/58425499/147/Poisson-Pareto-Burst-Process

"The size of any of these bursts has a Pareto distribution, and each of them is
transmitted at fixed rate r"