[pim] IGP based mutlicast arch RE: pim minutes from IETF 92 in Dallas

[Lucy yong <[e-mail]>]

Hi,



During Dallas meeting, the presentation of draft-yong-pim-igp-mutlicast-arch triggered a lot of discussions/debates as shown in the meeting minutes. Co-authors would like to address these concerns on the mailing list again so we can have further discussions in a broader community. Since several questions are similar but were asked in different ways, we summarize them into the following questions:



1.  What is the relationship between IGP based multicast and M-OSPF (RFC1584)?

*         IGP based multicast and M-OSPF has one common, i.e., use one protocol to support both unicast and multicast routing.

*         M-OSPF (RFC1584) specified one solution that is the source rooted distribution tree for multicast delivery in 20 year ago when there was no multicast requirement due to, obviously, lacking of applications.

*         M-OSPF solution, i.e., (S,G) based distribution, can't apply to the Data Center that supports NVO3. We'll document the multicast requirement in Data Center in the next draft revision.

*         IGP based multicast has a primary use case for NVO3 Data Centers. Each multicast group uses a bidirectional rooted tree for multicast packet delivery and both sources and receivers are the member of the group. Such solution meets NVO3 DC multicast requirements.

*         The solution draft is different from RFC1584 in terms of algorithm and protocol extension. Since we have implemented this in IS-IS first, prefer working on IS-IS based solution first, then OSPF.

*         M-OSPF was done 20 year ago. Since then, IGP has enriched many properties that could be beneficial for multicast delivery as well, such as traffic engineering, fast re-route and loop-free convergence, etc. This is a new area to explore. IGP multicast architecture intends describing general architecture and architecture components, rather than describing a specific solution. A solution will be a separate draft.



2.  Is this just like PIM but implemented by IGP protocol? Why do it again?



*         Answer is Yes and No. There are several PIM based multicast delivery methods that apply to intra IGP and inter IGP. One IGP based multicast solution we are working on is like PIM bidir method (RFC5015) in theory, but just works in intra IGP. This solution targets for the Data Centers where NVO3 is deployed. For this use case, our solution is better than PIM bidir: 1) enable infrastructure network self-establishment; 2) meet the requirement of supporting both unicast and multicast routing in DC network; 3) save the multicast convergence time. The benefit of using one protocol is highly desired by Data Center operators due to their requirement, e.g., reducing operating cost, simplifying network management/provisioning effort, and the "automation", i.e., when the IP network is brought up, both unicast and multicast routing engines are running.

*         There are already similar deployments such as TRILL (RFC5556) and IP fabric path in DCs. They both uses single protocol to handle unicast and multicast routing, where the operating experience and benefit have greatly encouraged the single-protocol vision, especially in Data Center networking environment.

*         IGP bidir tree solution is different from PIM bidir tree solution because IGP protocol and PIM protocol are different, e.g., group membership announcement mechanism is different. Our IGP multicast solution draft will have the details.

*         Improving/simplifying infrastructure network is a key requirement for Data Center network. We need to think of it from network perspective, not at individual protocol view.



3.  Concern on IGP Group Membership Flooding:



*         Yes, IGP floods the group membership while PIM does not. We believe that for some use case, such behavior is acceptable and other benefits of it comparing to PIM remains. In addition, some optimization at edge routers to minimize the flooding can be done in DC use case.

*         The flooding makes IGP multicast convergence faster than PIM solution.

*         In Data Center environment, the membership subscription is static in nature, i.e., the re-flooding due to new member's joining or existing member's leaving is rare. Therefore, the concern on the flooding is unnecessary.

*         We're currently conducting comparison between IGP-based multicast routing and PIM-based multicast routing in terms of performance, convergence, volume of protocol messages, etc. We'd share the result with the community once the data is ready.


4.  Inter-AS support?



*         IGP based multicast focuses on intra-AS use cases that, we believe, have huge market as the trend of network virtualization.

*         Our solution can support single area and multi-areas, and this covers the current Data Center networking environment well. However, inter-AS support can be added at later time if there is a requirement.


5.  Need to see and evaluate the solution.



*         We will upload a bidir tree solution based on IS-IS protocol first, which is similar to the draft-yong-isis-ext-4-distribution-tree-02, since we had an implementation.

We like to hear people's thoughts/suggestions on these and get people support for PIM WG to work on this development work.



Thanks,

Lucy on behalf of the co-authors of draft-yong-pim-igp-mutlicast-arch



-----Original Message-----
From: pim [mailto:[e-mail]] On Behalf Of Stig Venaas
Sent: Monday, April 13, 2015 12:37 PM
To: [e-mail]
Subject: [pim] pim minutes from IETF 92 in Dallas



Hi



The minutes have finally been posted. They are available at http://www.ietf.org/proceedings/92/minutes/minutes-92-pim



Stig



_______________________________________________

pim mailing list

[e-mail]<mailto:[e-mail]>

https://www.ietf.org/mailman/listinfo/pim