<!doctype html public "-//w3c//dtd html 4.0 transitional//en">
<html>
ROBERTS ,
<p>thanks for your comment.
<p>But, why do you think OMP <font size=-1>do nothing on maintaining
efficiency in overload and can arguably lead to the spread of congestion?
As I think OMP direct traffic based on local routing computation,
the exact problem is</font>
<br><font size=-1>with its dispersion method (hash or random ).</font><font size=-1></font>
<p><font size=-1>And , you means those elastic traffic as traffic
without strict qos requirement ? ( I'm sorry if I'm too new to this )</font>
<br><font size=-1></font>
<br><font size=-1></font>
<br><font size=-1></font> <font size=-1></font>
<p><font size=-1>Regards</font>
<br><font size=-1></font>
<blockquote TYPE=CITE>
<p><font size=-1>While it is true that the current connectionless IP architecture
offers limited possibilities for flow blocking, it may be argued that the
introduction of admission control would be a desirable development, notably
to allow intelligent traffic aware routing.</font>
<p><font size=-1>Congestion can be defined as a situation where traffic
demand (flow arrival rate x average flow size) exceeds capacity. In this
case, the number of flows in progress continues to increase while their
realized throughput tends to zero. This phenomenon is analysed in "Statistical
bandwidth sharing: a study of congestion at flow level" by S. Ben Fredj
et al. (<a href="http://www.acm.org/sigs/sigcomm/sigcomm2001/p9.html" TARGET="_blank">http://www.acm.org/sigs/sigcomm/sigcomm2001/p9.html</a>).</font>
<p><font size=-1>If a link is congested it would be preferable to reroute
flows over an alternative path. A lot of work has been performed on routing
strategies under the assumption that flows can be assimilated to constant
rate streams (or circuits) and admission control is performed in the control
plane. Dynamic traffic aware routing is generally regarded as preferable
to fixed path routing. In the Internet, routing is essentially fixed path
(flows cannot choose their path depending on current load) and this can
lead to bad quality and inefficiency if that path happens to be overloaded.</font>
<p><font size=-1>Some work on adaptive flow level routing for elastic traffic
is reported in papers on the site <a href="http://www.enst.fr/~oueslati/publienglish.html" TARGET="_blank">http://www.enst.fr/~oueslati/publienglish.html</a>.
An essential component is implicit admission control (realized in the data
plane) as discussed in papers on the same site.</font>
<br>
<p><font size=-1>OMP (and similar approaches proposed in MPLS Traffic engineering)
attempt to adapt load sharing between a number of possible paths and would
avoid the problem of the overloaded fixed path. They do nothing,
however, to maintain efficiency in overload (demand greater than the sum
of available capacities) and can arguably lead to the spread of congestion.</font>
<p><font size=-1>Jim Roberts</font>
<br><font size=-1></font> </blockquote>
<pre>--
Jing Shen
**********************************************************************
* The SunShine of life is made up of very little beams which is *
* bright all the time *
**********************************************************************</pre>
</html>