[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: Regarding the DiffServ-TE drafts(lefaucheur,boyel,kompella,as h,bi tar)
Some more thoughts (personal opinions) ...
1) We should separate the issue of advertising the
bandwidth associated with different diffserv classes
from -preemption priority advertisement.
->Eliminating the pre-emption related advertisements
reduces the scalability issues to a great extent.
->Now we are dealing with a traffic engineered network
,where the administrator wants more control and
predictability over the TE-LSP being created in his
I don't see, why the administrator would configure
his network in such a way that one of the LSPs he
established gets torn down automatically, 30 minutes
later, by another TE-LSP (of his own creation).
Furthermore, we are not creating "UNI" LSPs, which
gets created and deleted frequently and un-predictably.
->even if there are cases, where one really needs
pre-emption, we can address this issue via a separate
2) As I mentioned before, in my opinion a 1:1 approach
(diffserv class: bw adv), is probably the better approach.
The [kompella] drafts proposes 1:1 mapping. However, it
overloads the priority with the diffserv class bandwidth
info, there by tightly coupling the per-diffserv-class
bandwidth info with a priority. This has several limitations
a) user may not want to define a strict "priority"
between the diffserv classes.
b) limits the number of class to 8.
It is a good idea to limit the number of classes, but
this can be handled by configuration.
What about a modified approach that uses the concepts
outlined in the [kompella] drafts, but encoding in the
style of [bitar] -explicit PHB-ID . An additional MIB
may help the administrator to further limit the amount
of advertisement by:
a) limiting the number of classes for which the
BW information has to be _explicitly_ advertise
b) the frequency of advertisement
c) a possible threshold, which decides whether
the amount of change in the reserved bandwidth
should cause the initiation of the new per-class
Do you see any problem with this proposal ?
From: Kireeti Kompella [mailto:email@example.com]
Sent: Saturday, September 08, 2001 12:44 AM
To: firstname.lastname@example.org; email@example.com
Subject: RE: Regarding the DiffServ-TE
drafts(lefaucheur,boyel,kompella,as h,bi tar)
> The two drafts you requested are
You forget draft-kompella-tewg-bw-acct-00.txt, which by trading off
the number of priorities per class for the number of classes keeps
the size of IGP updates and the TE database small, between 1/4 and
1/3 the size of the LeFaucheur approach.