The MPLS-OPS Archive

Cell Relay Retreat>MPLS-OPS Archive>month:2004-May> msg00043



[Date Prev][Date Next][Thread Prev][Thread Next]  
  [Date Index][Thread Index][Author Index][Subject Index]

RE: some questions

  • From: "M. ELK" <elkou141061@hotmail.com>
  • Date: Sat, 08 May 2004 08:28:36 +0000
  • Cc: mpls-ops@mplsrc.com
  • Resent-Date: Sat, 8 May 2004 04:51:43 -0400
  • To: sthaug@nethelp.no
  • X-OriginalArrivalTime: 08 May 2004 08:28:36.0455 (UTC) FILETIME=[750B7F70:01C434D6]
  • X-Originating-Email: [elkou141061@hotmail.com]
  • X-Originating-IP: [57.250.229.136]
  • X-Sender: elkou141061@hotmail.com


Steinar

For the need of modeling

Sorce : The Measurement Manifesto Paper ( George Varghese et al )
( very nice doc addrssing the need to add some counters in the router to 
help
  calculate the traffic matrix )

Quote

If  even half the attention to 'rocket science traffic modeling" were 
devoted to how
to estimate a reasonable ingress-egress traffic matrix , network 
engineers,particularly of large clouds,
would find their job substantially easier .

- Dennis Ferguson - 1996 ISMA .

Unquote


my guess , Modelling could not be generalized . Every provider need to have 
the tools
to measure his network and try to find the patern specific to his network .

For the burst :

As fas as i understand this paper " Wide-Area Internet Traffic Patterns and 
Characteristics
(extended version) , Kevin Thompson et al , Nov/Dec 97 " from MCI .

Figure 3a,3b,3c,3d . show that the traffic is not so bursty ( measured at 
the core ) as we tend to thinck it is . Traffic show a pattern during the 
day and along the week .

Appreciate U comments .

Brgds



>The aggregated traffic *is* quite like white noise - in the sense that
>packet arrival times are random. The important point here is not that there
>are no peaks to the traffic, but that, as the number of aggregated streams
>increases, the ratio of the size of the peaks compared to the total amount
>of traffic gets smaller, so the flow is "smoother". Basically, you get a 
>big
>fat flow of data, with a few small spikes on top of it. Fractal traffic
>profiles do NOT behave like this. The peak size stays the same relative to
>the flow of data. Add lots of fractal traffic streams together and you get
>one large flow of traffic, with very big peaks and troughs.

It might be a good idea to read some of the influential papers in this
field, for instance Leland & al: "On the Self-Similar Nature of Ethernet
Traffic":

	http://citeseer.ist.psu.edu/leland93selfsimilar.html

Steinar Haug, Nethelp consulting, sthaug@nethelp.no

_________________________________________________________________
Add photos to your e-mail with MSN 8. Get 2 months FREE*. 
http://join.msn.com/?page=features/featuredemail

-------
The MPLS-OPS Mailing List
Subscribe/Unsubscribe:  http://www.mplsrc.com/mplsops.shtml
Archive: http://www.mplsrc.com/mpls-ops_archive.shtml