PANRG Agenda IETF 113
When: 13:00-14:00 CET Thu March 24th, 2022
Where: Meetecho
Chairs: Jen Linkova and Brian Trammell
Minutes Taker: Gorry Fairhurst, Carsten Bormann
Jabber Scribe: TBA
Jabber: panrg@jabber.ietf.org
RFC9217 (current open questions) published
draft-irtf-panrg-path-properties active RG document
Brian: SCION side meeting Tue; will need to find out how to engage
with the IRTF/IETF.
Spencer: I think we should announce this meeting really soon, so we
have opportunity to better prepare.
Cyril presents updates (host -> endpoint, ...)
Brian Trammell: I quite like the endpoint definition. This seems
sufficient.
Discuss last-call on the list...
Show-of-hands:
So what is missing?
Tianji Jiang: used "domain", is that term causing
conflict/ambiguity?
Cyrill: generic identifier, needs to be defined in terms of specific
routing protocol
Mohamed Boucadair_web_183 (in chat): I'm not sure to get the
comment. This is a generic abstract of a specific network permiter an
area can be an example of a domain for IGPs 1
How to use proxies
PEPs
are not strictly evil, good intentions, terrible outcomes
so QUIC does not work as well over satellite
retrofit proxies into QUIC?
ossification was due to PEPs being transparent; need to make assumptions
about headers that weren't justified
Suggestion:
separation of concerns: "sidecar"
opt-in
some compromise of anonymity is part of the contract
minimize changes to main protocol
ossifying the sidecar would lead to losing performance advantage worst
case
PEP as use case for sidecar
ACK hashes over transport
could be piggybacked using UDP options
using QUIC for exposition, could be TCP or SCTP
Example 1: link with fluctuating capacity (mmwave)
server picks service choice
SC gives QUIC an SC notification
CC requires control loop, needs to be on client side as well
Example 2: WiFi AP acks on behalf of host
server chooses to accept SC acks as client acks (but leave data in send
buffer)
PEPs sends hashes as ACKs
client probably should know that it doesn't need to ACK as much.
There is future research: how to limit SC ACK overhead;
rendezvous with SC proxy via sending hashes to sender
sender-side SC needs to "trust" proxy, which would need to guess hashes
Path change: some form of negotiation (hand-over)
Stuart Card (in chat): +1 to this sidecar idea
Carsten Bormann (in chat): This is Interesting stuff. (And I'm
already thinking how to put LOOPs in here :-)
Chris Box (BT): I like the idea.
Jake Holland: Ideas how to discover?
Marcus Ihlar: Ericsson is doing similar stuff; Light-weight PEP.
Mohamed Boucadair_web_183 (in chat): Is this another way to do
LOOPS?
Stuart Card_web_674: +1 to this sidecar idea
cabo: This is Interesting stuff.
Stuart Card_web_674: I am a long time PEP user/developer.
cabo: (And I'm already thinking how to put LOOPs in here :-)
Mohamed Boucadair_web_183: it is already some sort if LOOPS,
Cartsne ;-)
Stuart Card_web_674: ESP is essential but precludes what I have
always done.
Brian Trammell_web_735: it's like LOOPS + PLUS
Mohamed Boucadair_web_183: It isn't PLUS as the initial signal not
touched. No? this is more OOB
No questions.
Chairs expect an Interim meeting, and will announce to the list. There
is likely to be a request for a meeting in Philadelphia.