Hi Adrian,
On Jul 20, 2006, at 1:17 PM, Adrian Farrel wrote:
Hi,
This thread went suddenly very quiet in response to Igor's question
which is a shame because it is a good question.
When you are doing a path computation how do you know why you can't
reach the destination with sufficient bandwidth and low enough delay?
Is it because there is not enough bandwidth on the low delay path?
Or is it because there is too much delay on the high bandwidth path?
The issue becomes unmanageably complex when there are many
constraints, and is particularly difficult when those are absolute
not relative constraints.
Of course, it is always possible to make some guesses, but usually
these are based on varying the constraints to see what could be
achieved. The choice of which constraints to vary, by how much, and
in what order is very suspect, and (as when we discussed constraint
relaxation) should be the subject of policy either at the PCC or
the PCE.
For example, a response that says "If you relaxed the required
bandwidth by 10% you could get a path" is no use to a PCC that MUST
have the bandwidth, but that would be happy to relax the delay
constraint by 90%.
I think it is ... there are many network designs in which a clear
constraint hierarchy can be defined. For example, bandwidth and
affinity (and of course the number of such example is quite large).
"You can relax the affinity constraint if a path with at least 90% of
the required bandwidth can be found". In that particular case, it is
quite easy to specify an algorithm capable of satisfying such
request. There are also some cases where the issue is NP-hard (most
of them) but the number of situations where hierarchical constraint
relaxation would be useful is not negligible.
Note that we're discussing about constraint relaxation here for which
there was a decision to tackle this issue in a separate ID.
Cheers.
JP.
Adrian
----- Original Message ----- From: "Igor Bryskin" <[EMAIL PROTECTED]>
To: "JP Vasseur" <[EMAIL PROTECTED]>; "LE ROUX Jean-Louis RD-CORE-
LAN" <[EMAIL PROTECTED]>
Cc: <[EMAIL PROTECTED]>
Sent: Monday, July 17, 2006 5:03 PM
Subject: Re: [Pce] P & I flags
JP.
IB>> This is the whole my point. If there is a set of
mandatory constraints,
there is no way for a PCE to tell because of which particular
constraint(s)
the computation has failed,
Why, if it is clever enough to detect a blocking constraint?
I agree with JL here. There *are* many ways to figure out which
constraints could not be satisfied, in which case indicating this
information to the PCC is quite useful.
IB>> I am just curious. Could you (or anybody) describe just one
of the ways
how PCE can figure out which of mandatory constraints caused the path
computation to fail?
Thanks,
Igor
_______________________________________________
Pce mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/pce
_______________________________________________
Pce mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/pce
