|
Subject |
Re: Why do we need topology exchange? |
|
From |
Harvey Newman <Harvey.Newman@xxxxxxx> |
|
Date |
Fri, 15 Oct 2010 07:23:56 -0700 |
HI All,
My opinion is - you can abstract what you need if you do not
want to divulge your full topology, but eventually a lot of information
will be needed.
In the early phases just stitching up a few circuits should clearly be
the focus. Later, assuming success, the resources for circuits will
be limited so detailed choices, allocations to make efficient use
of the resources in the presence of multiple requests, and scheduling
will follow. Given the status of the LHC experiments as likely main
users, with many sites having large needs and the LHC Computing
Models changing, the requirements will ramp up quickly in complexity as well
as capacity soon after we get started.
Regards
Harvey
On 10/15/2010 6:19 AM, Victor Reijs wrote:
Hello John and Freek,
John Vollbrecht wrote:
The rest of topology I am not sure about. I would think it a good
thing, but I also think it would take a lot of work and be hard to
maintain. Is this what you mean?
In some way I like the idea we had in AutoBAHN: an abstracted topology
to determine the path.
An abstracted topology does not hold much detail just enough to
determine the possible path (I am see this as with human: when they
talk about a project they more or less have the physical connectivity
in their had.
The next step is are higher network protocol levels compatible.
This is the very first part of path finding (the net graph, which can
be in one or two stages).
And than can we make them work (Stitching Framework).
And then implement it (Signaling).
So A full topology is not needed in my humble opinion, but an abstract
(at least see a domain as a node).
All the best,
Victor
