Re: Multipath path issues for discussion at interim

[Lucas Pardue <[e-mail]>]

Hi all,

Please join us today, in a little under 3 hours to discuss the issues Mirja listed out in her email.

We'll be using Meetecho for remote participation: https://meetings.conf.meetecho.com/interim/?group=69f724f3-3d5d-438a-a532-ac06d539f842

Cheers
Lucas

P.S. There was some hiccup preventing the datatracker from linking to meetecho, the details in this email are confirmed as correct with the meetecho team.


On Fri, May 31, 2024, at 09:48, Mirja Kuehlewind wrote:
> Hi all,
> 
> we still have some more mayor design issues connected to the new explicit path ID approach that we would like to discuss at the interim, these are:
> 
> 1) Even odd/split:
> 
> - Main issue for discussion: #328 Designing odd/even path-id, or not. https://github.com/quicwg/multipath/issues/328
>   -> Note that the easiest solution would be to basically do nothing now because we can probably specify anything needed including the split with a new extension using a new transport parameter.
> 
> 2) Reconsider the path closing procedure
> 
> - Main issue for discussion: #313 When to CID expire if using Unique Path ID? https://github.com/quicwg/multipath/issues/313
>    -> the proposal is to change the closing procedure to retire CIDs immediately together with the path abandon (see issue for details)
> 
> - Related issue: #366 Should we require explicit Abandon after path time-out? 
> https://github.com/quicwg/multipath/issues/366
>    -> We already say SHOULD send abandon after time our or validation failure but it would probably make other issues easier if we do a MUST here to explicitly retire the CIDs related to this path ID
> 
> - Related issue: #318 Should path numbers be continuous? https://github.com/quicwg/multipath/issues/318
>    -> This is related because, having path IDs issued in order would make other issues probably a bit easier, so we should explicitly recommend that. However, we can probably not enforce it and therefore not rely on it.
> 
> 3) CID allocation strategy:
> 
> - Main issue for discussion: #338 Clarify how new CIDs are allocated https://github.com/quicwg/multipath/issues/338
>   - We now specify that the same path ID is used in both directions which could lead to a situation where each ends issues different path IDs and then no new oath can be opened. #318 is also related here because requiring the next in order path ID would avoid a mismatch but we still need to provide more recommendation about how many path IDs should be issued when.
> 
> - Related issue: #332 Should active_connection_id_limit be per path or per connection? https://github.com/quicwg/multipath/issues/332
>   -> This issue is important in order to decide if we need to signal max paths to the peer or not which impacts the CID allocation strategy regarding how many path IDs should be issued.
> 
> 
> Please have a look and comment on github before the interim. Sorry for the rather short notice!
> 
> See you at the interim on Monday!
> 
> 
>