Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

RE: RPR perf: My thoughts




Hi,

I agree with Bora on that we can get converged simulation results for
self-similar traffic although it may be a little bit slow. What I don't
agree is that, I think self-similar phenomena may show up at both congested
and uncongested situations. Both Bellcore and Paxon's results were meassured
at various times and loads for a quite long time period. One explaination
(by Corovella) is that self-similar is caused by original file size
distribution which shows heavy tail (e.g. files in UNIX systems). Clearly
file size distribution doesn't depend on congestion or not.

chang

> -----Original Message-----
> From: owner-stds-802-rprsg@xxxxxxxx
> [mailto:owner-stds-802-rprsg@xxxxxxxx]On Behalf Of Bora Akyol
> Sent: Thursday, January 25, 2001 6:13 PM
> To: Shahid Akhtar
> Cc: Harry Peng; khaledamer@xxxxxxx; Changcheng Huang (E-mail); Reflector
> RPRSG (E-mail)
> Subject: Re: RPR perf: My thoughts
>
>
>
> Shahid
>
> It is not wise to say that simulations don't converge when one
> uses self-similar traffic. There is numerous ways of generating
> self-similar traffic traces that are usable and result in
> repeatable experiments.
>
> Also, what has been shown so far is that TCP/IP traffic when not
> facing a hard limited bandwidth requirement exhibits
> self-similar behavior.
>
> The starts of individual "sessions" can be explained perfectly
> well by Poisson behavior.
>
> What has not been shown is what happens to TCP/IP traffic that
> is going through a hard-limited (congested) backbone.
>
> Bora
>
>
> >>>>> "Shahid" == Shahid Akhtar <sakhtartx@xxxxxxxx> writes:
>
>     Shahid> Khaled and Harry, Are we assuming self-similar data
>     Shahid> traffic for the simulations or some version of
>     Shahid> Poisson based traffic. Most internet traffic is TCP
>     Shahid> based which has been proven to be self-similar.
>
>     Shahid> The problem with simulating self-similar traffic
>     Shahid> (since it has infinite variance) is that the
>     Shahid> simulations do not converge.
>
>     Shahid> There are solutions via some analytical methods, but
>     Shahid> then the performance would be done mostly
>     Shahid> analytically.
>
>     Shahid> Regards,
>
>     Shahid> Shahid Akhtar Cyras Systems
>
>     Shahid> ----- Original Message ----- From: Harry
>     Shahid> <mailto:hpeng@xxxxxxxxxxxxxxxxxx> Peng To:
>     Shahid> khaledamer@xxxxxxx <mailto:khaledamer@xxxxxxx> Cc:
>     Shahid> Changcheng Huang (E-mail)
>     Shahid> <mailto:changcheng.huang@xxxxxxxxxxxxxxx> ;
>     Shahid> Reflector <mailto:stds-802-rprsg@xxxxxxxx> RPRSG
>     Shahid> (E-mail) Sent: Thursday, January 25, 2001 2:12 PM
>     Shahid> Subject: RPR perf: My thoughts
>
>
>     Shahid> Khaled: <<Performance Ad Hoc Proposal.ppt>>
>
>     Shahid> At last......As promised from last week's meeting.
>
>     Shahid> Attached is some charts that capture to what we
>     Shahid> discussed over lunch.  As you can see it addresses
>     Shahid> Raj's request for cut-through versus store and
>     Shahid> forward.
>
>     Shahid> You are welcome to modified the chart.  I'm
>     Shahid> basically requesting a need for a common test
>     Shahid> harness.  We need sponsors for the harness modules.
>
>     Shahid> I've also tried to summarize the different RPR MAC
>     Shahid> proposals.  We can seek each proposal's supporter to
>     Shahid> do the simulation for each proposal.
>
>     Shahid> Follow your guideline on traffic profile and metric
>     Shahid> we shall have a good platform for comparing the
>     Shahid> different proposals.
>
>     Shahid> I don't see how we can compare the results if we do
>     Shahid> not set a common simulation framework for each RPR
>     Shahid> node.
>
>     Shahid> Comments.....
>
>
>     Shahid> Regards,
>
>     Shahid> Harry
>
>     Shahid> <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0
>     Shahid> Transitional//EN"> <HTML><HEAD><TITLE>RPR perf: My
>     Shahid> thoughts</TITLE>
>
>     Shahid> <META content="MSHTML 5.00.2919.6307"
>     Shahid> name=GENERATOR> <STYLE></STYLE> </HEAD> <BODY
>     Shahid> bgColor=#d8d0c8> <DIV><FONT face=Arial size=2>Khaled
>     Shahid> and Harry,</FONT></DIV> <DIV>&nbsp;</DIV> <DIV><FONT
>     Shahid> face=Arial size=2>Are we assuming self-similar data
>     Shahid> traffic for the simulations or some version of
>     Shahid> Poisson based traffic. Most internet traffic is TCP
>     Shahid> based which has been proven to be
>     Shahid> self-similar. </FONT></DIV> <DIV>&nbsp;</DIV>
>     Shahid> <DIV><FONT face=Arial size=2>The problem with
>     Shahid> simulating self-similar traffic (since it has
>     Shahid> infinite variance) is that the simulations do not
>     Shahid> converge.  </FONT></DIV> <DIV>&nbsp;</DIV>
>     Shahid> <DIV><FONT face=Arial size=2>There are solutions via
>     Shahid> some analytical methods, but then the performance
>     Shahid> would be done mostly analytically.</FONT></DIV>
>     Shahid> <DIV>&nbsp;</DIV> <DIV><FONT face=Arial
>     Shahid> size=2>Regards,</FONT></DIV> <DIV><FONT face=Arial
>     Shahid> size=2></FONT>&nbsp;</DIV> <DIV><FONT face=Arial
>     Shahid> size=2>Shahid Akhtar</FONT></DIV> <DIV><FONT
>     Shahid> face=Arial size=2>Cyras Systems</FONT></DIV>
>     Shahid> <BLOCKQUOTE style="BORDER-LEFT: #000000 2px solid;
>     Shahid> MARGIN-LEFT: 5px; MARGIN-RIGHT: 0px; PADDING-LEFT:
>     Shahid> 5px; PADDING-RIGHT: 0px"> <DIV style="FONT: 10pt
>     Shahid> arial">----- Original Message ----- </DIV> <DIV
>     Shahid> style="BACKGROUND: #e4e4e4; FONT: 10pt arial;
>     Shahid> font-color: black"><B>From:</B> <A
>     Shahid> href="mailto:hpeng@xxxxxxxxxxxxxxxxxx"
>     Shahid> title=hpeng@xxxxxxxxxxxxxxxxxx>Harry Peng</A> </DIV>
>     Shahid> <DIV style="FONT: 10pt arial"><B>To:</B> <A
>     Shahid> href="mailto:khaledamer@xxxxxxx"
>     Shahid> title=khaledamer@xxxxxxx>khaledamer@xxxxxxx</A>
>     Shahid> </DIV> <DIV style="FONT: 10pt arial"><B>Cc:</B> <A
>     Shahid> href="mailto:changcheng.huang@xxxxxxxxxxxxxxx"
>     Shahid> title=changcheng.huang@xxxxxxxxxxxxxxx>Changcheng
>     Shahid> Huang (E-mail)</A> ; <A
>     Shahid> href="mailto:stds-802-rprsg@xxxxxxxx"
>     Shahid> title=stds-802-rprsg@xxxxxxxx>Reflector RPRSG
>     Shahid> (E-mail)</A> </DIV> <DIV style="FONT: 10pt
>     Shahid> arial"><B>Sent:</B> Thursday, January 25, 2001 2:12
>     Shahid> PM</DIV> <DIV style="FONT: 10pt
>     Shahid> arial"><B>Subject:</B> RPR perf: My thoughts</DIV>
>     Shahid> <DIV><BR></DIV> <P><FONT face=Arial
>     Shahid> size=2>Khaled:</FONT> <BR><FONT color=#000000
>     Shahid> face=Arial size=2>&lt;&lt;Performance Ad Hoc
>     Shahid> Proposal.ppt&gt;&gt; </FONT></P> <P><FONT face=Arial
>     Shahid> size=2>At last......As promised from last week's
>     Shahid> meeting.</FONT> </P> <P><FONT face=Arial
>     Shahid> size=2>Attached is some charts that capture to what
>     Shahid> we discussed over lunch.</FONT> <BR><FONT face=Arial
>     Shahid> size=2>As you can see it addresses Raj's request for
>     Shahid> cut-through versus store and forward.</FONT> </P>
>     Shahid> <P><FONT face=Arial size=2>You are welcome to
>     Shahid> modified the chart.  </FONT><BR><FONT face=Arial
>     Shahid> size=2>I'm basically requesting a need for a common
>     Shahid> test harness.</FONT> <BR><FONT face=Arial size=2>We
>     Shahid> need sponsors for the harness modules.</FONT> </P>
>     Shahid> <P><FONT face=Arial size=2>I've also tried to
>     Shahid> summarize the different RPR MAC proposals.</FONT>
>     Shahid> <BR><FONT face=Arial size=2>We can seek each
>     Shahid> proposal's supporter to do the simulation for each
>     Shahid> proposal.</FONT> </P> <P><FONT face=Arial
>     Shahid> size=2>Follow your guideline on traffic profile and
>     Shahid> metric we shall have a good platform for comparing
>     Shahid> the</FONT> <BR><FONT face=Arial size=2>different
>     Shahid> proposals.</FONT> </P> <P><FONT face=Arial size=2>I
>     Shahid> don't see how we can compare the results if we do
>     Shahid> not set a common simulation framework</FONT>
>     Shahid> <BR><FONT face=Arial size=2>for each RPR
>     Shahid> node. </FONT></P> <P><FONT face=Arial
>     Shahid> size=2>Comments.....</FONT> </P><BR> <P><FONT
>     Shahid> face=Arial size=2>Regards,</FONT> </P> <P><FONT
>     Shahid> face=Arial size=2>Harry
>     Shahid> </FONT></P></BLOCKQUOTE></BODY></HTML>
>