dean
Dean Cheng (dcheng) wrote:
Igor, Dimitri,
Please see some response to your questions below.
Dean
-----Original Message-----
From: dimitri papadimitriou [mailto:[EMAIL PROTECTED]
Sent: Friday, October 07, 2005 11:13 AM
To: [EMAIL PROTECTED]; Dean Cheng (dcheng)
Cc: [EMAIL PROTECTED]
Subject: Re: [Pce] Choosing the PCE protocol
[EMAIL PROTECTED] wrote:
Dean,
See my comments in-line.
Igor
As for COPS, its application environment is somewhat similar to that
of PCE but has some problems (if used for PCE applciation)
due to much optimization around its specifics such as:
1) The PEP (client) is the one required to initiate the
connection with a PDP (server). Whereas for PCE
application, either PCE (server) or PCC may initiate
the connection.
IB>> Why would a PCE want to initiate a connection with a PCC?
more generally i am not sure to understand why "PC" elements
would initiate connections ... are they not meant to "compute" paths ?
The "connection" I meant a transport (say TCP) connection if used. The
current PCE architecture does not make any assumption as who should or
should not initiate the connection. While it may be easier to agree
that why a PCC wants to initiate a connection with a PCE, there would
be deployment/application scenarios where a PCE initiate the
connections, and the PCEP does not prevent from doing that.
which scenarios ?
We could discuss such cases in Vancouver if you will.
i am ok for discussion but i am also in favor then to postpone this
argument until there is a common understanding here
2) There requires a synchronization between PEP and PDP for
the policy application with states maintained on both places.
For PCE application, the PCE server is only required to
perform a one-time service for each LSP, without the need
maintain any subsequent states.
IB>> Even if all this is true, why is it a big deal?
there is also a need to discuss what are these states and what
they represent in the context of this application (PCE) before
claiming that they are impacting one implementation and not the other
This may not be a big deal, but when PCEP is used, there is no need
to worry about the deal, nor the protocol backward-compatibility. No
need to have the switch/case statement in the existing code if any
either.
indeed, PCEP is a dedicated new protocol so there is no backward
compatibility issue possible
but does it prove something about the intrinsic quality of PCEP vs COPS
? i think the issue should better be positioned when the whole landscape
will be available
3) It requires some work to support a PDP functions as a PEP
and communicates with another PDP. PCEP supports this model
fully.
IB>> Functions of PDP and PEP are orthogonal and surely could be
physically co-located indeed
Again, this may not be a big deal, but PCEP explicitly supports the
two modes with well defined functions and operations.
the document is still an individual submission so i would not presume
about the level of readiness (of any) particular solution - once the WG
will converge one should expect several cycles to reach some stability
4) COPS is a server-client protocol, but with code point
heavily dedicated for a specific application. It does not
serve general purpose if any, and to support PCE application,
there would be a large number of extensions, thus defeating
the purpose of reusage.
IB>> PCIM is designed to be easily expanded for application
(read PCE) specific needs
Probably the amount of the work to add extensions to the COPS is a
large part of the PCEP already, and you only re-use a server-client
communication mechanism that is a relatively small piece. And you
need to do some filtering on the COPS code avoiding those not applied.
this underlines the need to come out with a more complete document for
assessing these statements
As a result, I'd favor PECP as a PCE-specific protocol; it is much
cleaner and also light-weighted.
there is very little material/argumentation as part of this
e-mail that allows to come to such conclusion
by the way i would like to ask what is meant here by "cleaner"
? and how it is possible to come to the conclusion that PCEP
is lightweighted by the argumentation provided here above ?
That means as a dedicated PCE communication protocol, it is designed
solely for the PCE application, without carrying any non-applicable
or useless protocol element.
so by cleaner you meant "dedicated" (the latter sounds obvious since
referred to as PCEP) but i am unclear on why (even if your feeling is
that one would have to cope with non-applicable elements using COPS)
this makes PCEP lightweighted ? as said in a previous thread the light
vs heavy argument should be used iff there is a common understanding on
what these terms exactly mean
Dean
-----Original Message-----
From: [EMAIL PROTECTED]
[mailto:[EMAIL PROTECTED] On Behalf Of Dean Cheng (dcheng)
Sent: Monday, October 03, 2005 3:30 PM
To: Adrian Farrel; [EMAIL PROTECTED]
Subject: RE: [Pce] Choosing the PCE protocol
Hi,
I don't think HTTP is a good candidate for the PCE application and
here are some weakness one could think of, if it used for PCE
application:
1) The HTTP is specifically designed for communications between
web server and its client, or applications in that nature.
Although it is possible to add extensions to support PCE
application, it would require a considerable amount of work.
That is not worth of the effort given the much smaller and
narrower application scope of PCE.
2) The payload of HTTP messages (both request and response)
is in form of byte stream, which is unstructured comparing
to the PCEP proposal, where the content is structured with
objects.
For the PCE application, the object based format is more
efficient in terms of processing.
3) The HTTP does not have unsolicited messages such as
Notification
message and Error message that are defined by PCEP.
4) PCE architecture allows a PCE functions as a server
but also as
a client communicating with another PCE server. In both cases,
the same protocol is used. PCEP supports this scenario. There
requires some work to make HTTP behavior as such.
Dean
-----Original Message-----
From: [EMAIL PROTECTED]
[mailto:[EMAIL PROTECTED] On Behalf Of Adrian Farrel
Sent: Thursday, September 29, 2005 7:27 AM
To: [EMAIL PROTECTED]
Subject: [Pce] Choosing the PCE protocol
Hi,
Three candidate protocols have been brought to my attention for us
to consider. These are (in no specific order):
- HTTP
- PCECP (as draft-vasseur-pce-pcep-02.txt)
- COPS
I would like to close the nomination period, and move on to
select our
protocol.
The best approach will be if supporters of the candidate protocols
indicate (briefly) why their proposed protocol is
suitable. It would
also be helpful if they were frank about what the weaknesses of
their proposal are.
And please - this is not an election, it is an interview process.
Thanks,
Adrian
PS The reason why JP is keeping quite on this is that he is an
author of draft-vasseur-pce-pcep-02.txt and so (quite rightly) is
not occupying the chair during this discussion.
_______________________________________________
Pce mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/pce
_______________________________________________
Pce mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/pce
_______________________________________________
Pce mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/pce
_______________________________________________
Pce mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/pce
.
.
_______________________________________________
Pce mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/pce
