Hi Dan,
On Jul 11, 2007, at 7:57 AM, Dan Li wrote:
Dear JP,
Thanks for your comments!
Sorry for the late response, please see inline words in Blue.
Best regards,
Dan
----- Original Message -----
From: JP Vasseur
To: Dan Li
Cc: JP Vasseur ; Raymond Zhang ; Nabil Bitar ; JL Le Roux ; Adrian
Farrel ; [EMAIL PROTECTED]
Sent: Tuesday, July 10, 2007 4:05 AM
Subject: Re: New draft related to BRPC
Hi Dan,
On Jul 7, 2007, at 6:29 AM, Dan Li wrote:
Dear the authors of BRPC,
We just published a new draft which we think is worth to be
reviewed by you:
http://tools.ietf.org/wg/pce/draft-xia-pce-hybrid-network-00.txt
The purpose of the draft is:
- This draft is primarily to explain a problem that we think is a
real deployment scenario;
- We believe that BRPC can be used to solve the problem;
- If the WG is in agreement, we would like to propose a few
paragraphs to be added to [brpc];
- If the WG would like to discuss this issue more, we would be
happy to present it in Chicago.
Please let us know what's your suggestion. We greatly appreciate
your time to look at this new draft.
As you know the whole point of using a Multi-PCE approach with BRPC
is to compute
the shortest inter-domain constrained path (in addition to diverse
paths, ... ).
Agree!
The case you're looking at in this ID (called Hybrid network) can
no longer guarantee that
you'll find the shortest path of course.
Agree!
Referring to your first example, the most downstream PCE will only
see the next hops
to the downstream domain. So even if by using BRPC between the
first set of domains
where you have cooperative PCEs you get the shortest path between
the source and
the set of entry boundary nodes of the first non-PCE domain, this
does not guarantee
you anything in term of path efficiency ...
The purpose of this ID is try to look at the possible solution
where the PCEs are not available in all the domains. By using BRPC,
we can get the shortest inter-domain path. So even we could not get
the shortest inter-domain path by applying BRPC to all the domains,
we still can get the less shortest inter-domain path by applying
BRPC to most domains. This is kind of best effort path computation.
But that does not help, since even in the case (PCE-domain)(PCE
domain)(non PCE domain) you still do not know whether the best
path up to the exit BR of the second domain is the one you should
use. It might just be the worst possible path depending on the path
from the entry BR to the destination in the last domain. This is
clearly non deterministic.
And of course the (non PCE-domain)(PCE domain)(PCE domain) or (PCE-
domain)(non PCE domain)(PCE domain) cases are even worse.
Furthermore, in that case, you can also not compute diverse paths for
the exact same reason (another motivation for using a Multi-PCE
based approach).
Furthermore since you get several paths, would you try to signal
two TE LSPs in which
case they may end up blocking each other ?
Only one TE LSP is signalled in this case. Since there are several
optional TE LSPs, one of the optional TE LSPs may be signalled if
the first one could not be established successfully.
OK so that does not give you any quantifiable gain.
Note that this comment applies to the first two cases of section 2.2
The case 3 of section 2.2 suggests to use Virtual Inter-AS TE
Links: please refer to the
discussions on the list with regards to TE aggregation: this option
has been examined
a number of times and has been rejected. Adrian and I still have in
our plate to document this.
Sorry I was not aware of the history of the Virtual Inter-AS TE
Links discussions. It will be very helpful if someone can give me
more details on this issue, thanks.
Stay tuned, we'll try to post an ID soon.
And of course, if you start to alternate a number of PCE enabled
and non PCE enabled
domains, the result gets not only extremely complex but more
importantly highly unpredictable
in term of path quality.
Maybe the realistic world is not so horrible, the most possible
scenario is that one of the PCEs is broken in multi-domain network.
In this scenario, in order to get as much optimal inter-domain path
as we can, I think we need try to use the BRPC in most domains
where the PCEs are still available. Based on this understanding, we
think it's necessary to clarify how to apply the BRPC to the
scenarios described in this I-D.
See unless you can quantify the potential gain, which you cannot in
this case, I prefer not to add another layer of complexity.
Makes sense ?
Thanks.
JP.
Thanks.
JP.
Best regards,
Dan Li & Hongmiao Xia
_______________________________________________
Pce mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/pce
