I think you might predict PCEP communication time as...
TCP connection setup
PCEP session establishment
PCEP request/response
You might approximate this with a web browser or any other TCP-based
application.
FWIW I would expect a simple CSPF to be computed in a few hundred ms at
absolute most (but probably in a less than a few tens of ms) and
communication time to be of the order of one second tops. YMMV.
Cheers,
Adrian
----- Original Message -----
From: "Meral Shirazipour" <[EMAIL PROTECTED]>
To: "Adrian Farrel" <[EMAIL PROTECTED]>
Cc: <[email protected]>
Sent: Tuesday, August 05, 2008 5:44 PM
Subject: Re: [Pce] PCE computation time
Hi Adrian,
For the PCE computation time, I was considering all that you mentioned
except
the compiler efficiency/software implementation dependencies.
Also when you say:
In general,
for this type of computation, we observe that a remote PCEP communication
takes far more time than the computation. (Times may vary according to
the
pre-existence of the PCEP session.)
are we talking "worse case" PCEP communication time in the order of
seconds?
minutes? ..?
I am assuming that:
PCE response time = PCEP communication time + PCE computation time
Then the total response time is some sort of a summation of the response
times
of all the PCEs in the PCE chain.
Any further comments would be greatly apperciated. I am trying to come up
with
"typical/realistic" PCE response times for the single LSP computation
case. I
do understand that this time could greatly vary depending on the criterias
you
mentioned (CPU power, etc.), but just knowing if in practice it is
typically in
the orders of seconds, minutes, hours would be of great help to me.
many thanks,
Meral
Now then, Meral,
I think computation times *might* be dependent on CPU power, loading,
compiler efficiency, software implementation, algorithm choice, network
complexity, etc.
Some observations suggest that computation for single LSPs with normal
constraints can be achieved using CSPF and there are studies recording
the
number of logical steps needed to perform such a computation. In general,
for this type of computation, we observe that a remote PCEP communication
takes far more time than the computation. (Times may vary according to
the
pre-existence of the PCEP session.)
Other observations note that some computation problems are quite hard and
need a few more cycles. In these cases, given the low power of CPUs on
NEs,
the communication time may be less significant.
Cheers,
Adrian
----- Original Message -----
From: "Meral Shirazipour" <[EMAIL PROTECTED]>
To: <[email protected]>
Sent: Tuesday, August 05, 2008 12:24 AM
Subject: [Pce] PCE computation time
> Hi,
> Does anyone have actual values for the min/average/max PCE
> computation
> times
> in a real network?
> I remember that the issue was discussed for the monitoring draft.
>
>
> Thanks,
> Meral
>
>
> _______________________________________________
> Pce mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/pce
>
_______________________________________________
Pce mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/pce
