p4-apps@lists.p4.org

P4 Applications Working Group

View all threads

Meeting notes from 1/16/2020

LJ
Lee, Jeongkeun
Tue, Jan 28, 2020 12:44 AM

Dear members,

We had the first meeting of this year on Jan 16th.

Attendees:

Barak G. (Mellanox), Ramesh S. (Cisco), Mukesh H. (VMware), Mickey S. (Barefoot/Intel), JK L. (Barefoot/Intel).

First topic was on having two dataplane telemetry specifications in P4.org and IETF.

Barak raised a few points

  • P4 INT and IETF IOAM share lots of similarity
  • most P4 meeting attendees also participate in IETF
  • customers may get confused, partially due to terminologies

Ramesh has been the main contributor to INT v2.0 features and he shared the following points

  • P4 WG provided timely feedback, enabling quick iterations between the dev team and WG members. He was able to reach the consensus and get a timely closure.
  • the P4 specs allow efficient and flexible implementation in report generation: e.g., raw report format and the Yang model
  • Supporting IOAM and INT together is not a big deal, needing a simple parser change.

JK brought up a couple of aspects, re: P4 Apps charter

  • Thanks to the programmability, the networking industry is moving towards more SW-driven. Having multiple APIs or SW solutions for the same goal (e.g., Thrift and gRPC for RPC) is common and encouraged, especially in the fast changing opensource world.
  • We should not change the INT header format and protocol semantics every often, but quick iterations and revisions in the speed of SW has been one of the goals in P4 Applications WG.

We discussed IETF's Informational RFC as another option for faster iterations; some mentioned that getting a presentation slot in IETF meetings (to be discussed for an approval) can be also challenging. If someone has more understanding on the Informational RFC approval process, that'd be great to share.

So far, both P4 INT and IETF IOAM have mostly focused on switch-driven telemetry use cases. NIC/host-driven use cases may require additional features and different semantics in INT. We decided to give it a try in the following meetings and see how the fast iterations will help such new use cases and applications.

Second topic was on the v2.0 PRs. Most open issues were discussed and resolved. We believe additional meeting is not required to close and merge the PRs. Follow-up discussions are happening in github. We will close most PRs this week, the remaining ones next week, and cut the v2.0 specs.

Next meeting: we planned to have the next meeting this Thursday but some key members have conflicts. We will move one or two weeks after.

Thanks,

JK Lee

Mukesh Hira

Dear members, We had the first meeting of this year on Jan 16th. Attendees: Barak G. (Mellanox), Ramesh S. (Cisco), Mukesh H. (VMware), Mickey S. (Barefoot/Intel), JK L. (Barefoot/Intel). First topic was on having two dataplane telemetry specifications in P4.org and IETF. Barak raised a few points * P4 INT and IETF IOAM share lots of similarity * most P4 meeting attendees also participate in IETF * customers may get confused, partially due to terminologies Ramesh has been the main contributor to INT v2.0 features and he shared the following points * P4 WG provided timely feedback, enabling quick iterations between the dev team and WG members. He was able to reach the consensus and get a timely closure. * the P4 specs allow efficient and flexible implementation in report generation: e.g., raw report format and the Yang model * Supporting IOAM and INT together is not a big deal, needing a simple parser change. JK brought up a couple of aspects, re: P4 Apps charter * Thanks to the programmability, the networking industry is moving towards more SW-driven. Having multiple APIs or SW solutions for the same goal (e.g., Thrift and gRPC for RPC) is common and encouraged, especially in the fast changing opensource world. * We should not change the INT header format and protocol semantics every often, but quick iterations and revisions in the speed of SW has been one of the goals in P4 Applications WG. We discussed IETF's Informational RFC as another option for faster iterations; some mentioned that getting a presentation slot in IETF meetings (to be discussed for an approval) can be also challenging. If someone has more understanding on the Informational RFC approval process, that'd be great to share. So far, both P4 INT and IETF IOAM have mostly focused on switch-driven telemetry use cases. NIC/host-driven use cases may require additional features and different semantics in INT. We decided to give it a try in the following meetings and see how the fast iterations will help such new use cases and applications. Second topic was on the v2.0 PRs. Most open issues were discussed and resolved. We believe additional meeting is not required to close and merge the PRs. Follow-up discussions are happening in github. We will close most PRs this week, the remaining ones next week, and cut the v2.0 specs. Next meeting: we planned to have the next meeting this Thursday but some key members have conflicts. We will move one or two weeks after. Thanks, JK Lee Mukesh Hira