PCE Working
Group Meeting Wednesday, March 25, 2015; 1:00-3:00 PM
1.
Introduction
1.1. Administrivia, Agenda Bashing
(chairs, 5 min)
- No change
requests.
1.2. WG Status (chairs, 20 min)
[25/120]
- Daniel
King to comments to the list of the I-Ds related to milestones.
- Dhruv
Dhody has some editorial updates related to the Service Aware I-D.
- Xian Zhang
thought it was useful for pending issues to be discussed during this IETF week
with the other authors in preparation for Last Call (aim to be prior to IETF 94).
- Oscar Dios
Gonzales had some cosmetic updates for his I-D.
2. Work in
Progress
2.1. PCEP Extensions for Segment
Routing (Jeff Tantsura, 10 min) [35/120]
draft-ietf-pce-segment-routing
- A request for clarification as to the length number supported for
SR-PCE-CAPABILITY. Presenter (Jeff Tantsura) clarified that length supported is
four.
- Concern raised that if the maximum SID depth in the Open message, which
effectively creates an implicit constraint on all queries that are sent over
the PCEP session, such that returned paths must not have a SID stack depth
greater than the MSD. This discussion
will be moved onto the list.
- Request received for an implementation section, which was acknowledged
by presenter as to be added.
- No Comments.
2.2. PCEP Extensions for SFC in SR
Networks (Jianjie You, 10 min) [45/120]
- An adoption request was delayed by Chairs as the I-D has a reference to
an individual document, so it is premature at this stage to adopt.
- Request at mic for applicability of I-D to Virtual Network Functions
(VNFs), as they currently lack capability to use SF Identifier (SF ID).
2.3. YANG Model for PCEP (Dhruv Dhody,
5 min) [50/120]
- Observed that an informal team was created to develop YANG as PCE
management tool. Suggestion that this decision and continued discussion might
want to be moved to the list.
- Comment received that the I-D scope is wide currently, and could be
expanded further as Stateful capabilities and others, are added. Presenter
mentioned if it becomes unmanageable the authors can break the I-D apart.
- Chairs declined a request for adoption as the work is still a little
premature.
2.4. PCEP Extensions for MPLS-TE
Auto-Bandwidth (Dhruv Dhody, 5 min) [55/120]
draft-dhody-pce-stateful-pce-auto-bandwidth
-
Need to remove code points from the document.
-
I will take this into consideration.
-
Has an option to include absolute thresholds
-
Is the Auto BW a management parameter? Take it offline.
Ina Scaling info from the sample interval. Concerns beyond small scale
deployments.
JP I designed auto-BW he dynamics, thousands of LSPs,
training/trending to anticipate the impact on the network. In theory its
interesting, practically it may have issues. Impact of 1000s of LSPs, using
auto BW.
Dhruv Some LSPs it may be important, not all LSPs. It is intended for a
stateful environment.
JP Have a discussion for scalability.
Ina two modes of operation. One where the LSP is created by the PCE.
Second, report BW back and PCE is resizes them.
Chris - Auto BW is fairly widely deployed in distributed mode adding
capability, it can only help.
2.5. PCEP Extensions for Resource
Sharing (Haomian Zheng,
5min) [60/120]
JP Few
items, a long long time ago. Impact on the network is significant, just because
it is 1-bit. Example fully diverse, different in costs. Notion of sharing may
require more
Lou Consider
ECMP LSPs, you have BW in the request and you want to support ECMP. You want to
allow protection sharing.
See Jabber
How is the PCC are already sharing resources?
draft-zhang-pce-resource-sharing
3. Next
Steps for PCE Scope and Charter?
3.1. Possible Candidate: Deterministic
Ethernet (Pascal Thubert, 15 min) [75/120]
3.2. Open Mic (all, 45 min) [120/120]
Lou Applicability in Decnet or 6tisch, then its DECNET but the
requirements are applicable to 6tisc.
Lou Work in CCAMP, including protection, in the context in CCAMP. Your
proposal is focused on centralised approach.
ADVA is detnet TDM? 6tisch is TDM, sequence of cells of TDM slots. We
also have x-axis for channels.10,000s of time slots. Maybe ODU model, aggregate
timeslots. IF we can model control plane point of view.
Pascal Software will be too slow. If the time slots are not used, they
can be used for other services.
Daniele Where is this work going to be progressed?
JP if you look at the architecture, they are using PCE but they may not
call it PCE.
Dhruv Why not use PCEP to the device?
Pascal It would be in combination with our UDP path setup.
Oscar Are these long live flows, an order of magnitude.
Pascal We need to design for different scenarios. If you scale you may
need multiple PCEs.
Oscar is there a data plane taking care of the issues.
Xian main application is sensor network
Adrian Last chance to tell you what to do. You need to gather round a
whiteboard and its probably a physical interim.
4. In case
time permits...
PCE as a
Central Controller (Dhruv Dhody)
draft-zhao-pce-pcep-extension-for-pce-controller
- Out of
time, did not present.