attached mail follows:
> Sorry in advance for receiving duplicate copies of this message
>
> I' m looking for mathematical models for TCP/IP sources, for every
> possible service (e.g. WWW, e-mail ....).
> The thing, that would more help me, were generated (and received)
> traffic patterns of a single TCP/IP source.
>
> Please send to me, through private mails, any references or results you
> can share.
Dr. Matarasso -
I am currently starting some research into accurate models for TCP
sources. I have not reached any conclusions yet, but I can give you
a list of some papers I have looked at so far.
1.) Walter Willinger, et. al., "Self-Similarity Through High-Variablility:
Statistical Analysis of Ethernet LAN Traffic at the Source Level,"
Proceedings of ACM SIGCOMM '95, pp. 100-113.
This papers starts with observations of aggregate network traffic and
explains how the self-similar properties of the aggregate traffic can
be attributed to individual ON/OFF sources which exhibit the Noah Effect.
It does not differentiate between WWW, email, or other types of sources.
2.) Vern Paxson and Sally Floyd, "Wide-Area Traffic, The Failure of
Poission Modeling," IEEE/ACM Trans. Networking, June 1995, pp. 226-244.
This paper looks at how individual telnet and ftp sources may be modeled.
It reaches similar conclusions as to the self-similar nature of aggregate
traffic and the large variance of burst sizes for individual source.
The telnet and ftp models have been integrated into a publicly available
network simulator called ns. There is more information on the simulator
at http://www-mash.cs.berkeley.edu/ns/
I do not yet have a good reference for WWW traffic models.
As I mentioned above, I started this research only a few weeks ago, so
I do not have a complete understanding of all of the problems relating
to modeling TCP sources. I would be interested in any additional information
you might have about this problem.
One final note: You mentioned using generated traces for TCP traffic. There
is a problem with using statically generated TCP traffic because TCP uses
a dynamic flow control mechanism. To accurately model TCP, it is necessary
to consider how the interaction of multiple flows will affect the flow
control mechanism. You cannot simply take multiple traces of TCP traffic
and combine them.
chuck
--------------------------------------------------------------------
Chuck Fraleigh | [email protected]
Multimedia Networking Group | http://www-leland.stanford.edu/~cjf
Stanford University |
This archive was generated by hypermail 2b29 : Mon Feb 14 2000 - 16:14:46 EST