INT-MX 'source-only' metadata embedded by source and reported by each node

[mickeyspiegel]

• In the INT-MX header, unlike other metadata, 'source-only' domain specific metadata is embedded in the packet after "DS Instruction" and "DS Flags". This 'source-only' metadata should be reported by each node including the source, transit, and sink nodes.

• In the Telemetry Report, the 'source-only' metadata may be reported in a truncated packet containing the INT-MX header, in domain specific extension data, or in "Variable Optional Domain Specific Metadata" with the corresponding bit set in "DSMdbits".

• In each of the "INT-MD Metadata Header Format" and "INT-MX Header Format" sections, consolidated the transit node processing text:

  • Moved text on INT-capable node limited in instructions or inserted metadata length, from Hop ML to transit node processing bullet.
  • Moved text on unknown Domain Specific ID from DS Instruction to transit node processing bullet.
  • In INT-MD, consolidated the text on INT-capable node limited in instructions or inserted metadata length with the text on a node that cannot process all required metadata, since they are redundant.
  • In INT-MD, clarified the order of metadata insertion in the transit node processing bullet.

• Reorganized INT-MX transit node processing text by splitting 2 options:

  1. Modify "RepMdBits" and "DsMdbits" in the Telemetry Report, after copying from INT-MX "Instruction Bitmap" and "DS Instruction" respectively, and recompute "MD Length".
  2. Copy INT-MX "Instruction Bitmap" and "DS Instruction" to "RepMdBits" and "DSMdbits" in the Telemetry Report, make minimal modifications as required, and adjust "MD Length". Made these two options the high level bullets, and then the detailed processing for unavailable metadata, reserved metadata, unknown "Domain Specific ID", and max exceeded become the inner bullets under each of the two options.

[mickeyspiegel]

The first two commits are reorganization of existing text. The third commit contains the technical changes.

[mickeyspiegel]

There are lots of tricky details regarding unknown Domain Specific ID and backwards compatibility implications. Pay attention when reviewing.

[mickeyspiegel]

After splitting the two main options for INT-MX transit node processing, there is a question whether the second option with padding is too complicated. Should it be deleted?

[jklr]

Thanks Mickey. I will review. I vote for deleting the second option with padding, for the sake of simplicity.

[jklr]

If we remove option #2, we can remove HopML from the MX header.

Q1. Can we also remove E and M bits and make them all reserved? Q2. how about adding 24b SEQ# in the freed-up space? SEQ# seems commonly needed by edge-to-edge and also some congestion control use cases.

[mickeyspiegel]

@jklr We can remove HopML if it is no longer used when forming Telemetry Reports.

I believe @rsivakolundu wanted to keep the E and M bit definitions when I asked a while ago.

I don't believe seq # should always be there. It makes sense from NICs, but it is difficult for switches to insert. With all the changes to allow source-only metadata, that seems like a perfect place for seq # to go.

[rsivakolundu]

I am ok with removing E and M bit and HOP-ML field and making all of them reserved. Cannot deal with using those reserved fields for anything at this stage.

[jklr]

sure. Let's remove E, M and HopML and keep them all as reserved bits.

[mickeyspiegel]

@jklr I threw an INT-MX with domain specific source-only metadata consisting of a Flow ID and Sequence Number into a separate example at the end. I also moved the INT-MD domain specific source-only example from Randy to the end of the Examples section instead of being stuck in an appendix.

[mickeyspiegel]

I will open a new PR for removing option #2 (padding), HopML, E, M, which will touch Telemetry Report as well as INT.