Hi all a couple more comments on top of Young's and Fabien's. See inline
below.
Best Regards
Greg
Young Lee wrote:
-- snip --
B) section "2.1.1. Nodes Send TE Info to all PCEs"
"As the number of PCEs grow we have scaling concerns"
Is it really a concern? How many PCEs do you think may be needed?
Young>> Each node will need to maintain session to each PCE with this
architecture and this could be burden to the nodes if there are "too many"
PCEs. If we are talking about 2-3 PCEs, the scaling concern may not be a
real issue.
--->> Agree, initially, this is a good place to start especially for
small WSONs.
-- snip --
D) Not sure what is the advantage of architecture 2 compared to architecture
1.
At first I thought the advantage was that the node only sent information to
one entity instead of N PCEs.
But since we also need a backup server for resiliency purpose, eventually it
seems to me both architecture
are equivalent.
Young>> Architecture 2 employs publish/subscribe functionality similar to
BGP route reflectors. P/S server needs not be a PCE although it can be
collocated with a PCE. We definitely need to consider resiliency of P/S
servers as you indicate.
Actually I think my problem is that I do not see what is the purpose of
having multiple PCEs for one intermediate
server.
I thought the goal of having multiple PCEs (in case 1) was for backup
purpose. This does not seems to be the case for case 2
since the backup is provided by having 2 intermediate server.
Young>> You may be right. We are exploring different possibilities in the
draft. We may elaborate advantages/disadvantages for each architecture.
---- Greg >> This case was trying to cover lots of PCE and the scaling
issue of the NEs talking to the PCEs.
E) section "3.2.2. Communication Protocols"
One idea:
Isn't it possible to meet the presented architecture by using IGP over
tunnels.
For case 1 for instance. In figure 1 the characters "|", "-", "/", or "\"
represents some (GRE) tunnels.
You run OSPF (or ISIS) over those tunnels so that there is an adjacency
between each node and the PCE in a dedicated OSPF (or IS-IS) area.
You can rely on regular OSPF/ISIS procedure to send TE information from node
to PCE.
The only OSPF/ISIS trick needed is at the PCE to prevent LSA received from
one node to be sent to other nodes.
I think this it is possible, without having any problem with legacy
OSPF/ISIS node though it needs to be check more deeply.
The advantage would be that it requires none or few protocol extensions.
One drawback is that it offers less flexibilities to target which
information are sent to the PCE and which are not.
Young>> Sounds a good choice. At this point, we are not proposing any
solution. This draft simply illustrates a set of possible communication
protocols that can implement the proposed architectures. We can add your
tunnels + IGP idea into the text. Ultimately, once this work is approved,
then we can think of implementation details.
-- Greg >> The new OSPF multiple-instance drafts could also help in this
area.
-- snip ---
--
===================================================
Dr Greg Bernstein, Grotto Networking (510) 573-2237
_______________________________________________
Pce mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/pce
