Thanks, Christian, I am happy to see what a minimalist design would look
like.

Some comments:
At the end Section 2 you discuss paths that have been validated, but not
received an ack-eliciting frame. But as PATH_CHALLENGE/RESPONSE are
ack-eliciting, what's the true distinction here?

3.1 The TP section should probably specify if it can/must be kept for 0RTT,
though as the frames are 1RTT specific the answer is obvious.

3.3 "Endpoints MUST NOT send PATH_STATUS frames for a Path Identifier until
the peer has acknowledged a PATH_IDENTIFICATION frame containing that
identifier, unless the frames are sent in the same QUIC packet." Otherwise,
we have messy states where PATH_STATUS arrives before PATH_IDENTIFICATION
and the receiver discards it, but the packet is acked and the sender has no
idea it's discarded. Alternatively, we could change the language about
discarding PATH_STATUS.

4.1 Senders that want to control this issue may do so

   by dedicating sub-ranges of packet numbers to different paths.

There be many dragons here! This would seem to violate the requirement
that packet numbers be sent in increasing order, which is a mechanism
that underlies several designs (e.g. key update).

Thanks again!

Martin


On Tue, Oct 20, 2020 at 11:47 PM Christian Huitema <[email protected]>
wrote:

> Since everybody seems intent on discussing Multipath, I just wrote this
> short draft that presents a minimalist design for adding Multipath support
> to QUIC. Basically, it puts the onus entirely on the sender. The receiver
> side would just require support for a couple of management frames to
> explicitly abandon or promote a path. The sender will have to track the
> relation between packet number and path, possibly manage ranges of packet
> numbers, manage the per-path congestion control, update the packet loss
> detection to take path delays into account, and of course chose an
> algorithm for scheduling packets to paths.
>
> -- Christian Huitema
>
>
> -------- Forwarded Message --------
> Subject: New Version Notification for
> draft-huitema-quic-mpath-option-00.txt
> Date: Tue, 20 Oct 2020 23:42:14 -0700
> From: [email protected]
> To: Christian Huitema <[email protected]> <[email protected]>
>
>
> A new version of I-D, draft-huitema-quic-mpath-option-00.txt
> has been successfully submitted by Christian Huitema and posted to the
> IETF repository.
>
> Name: draft-huitema-quic-mpath-option
> Revision: 00
> Title: QUIC Multipath Negotiation Option
> Document date: 2020-10-20
> Group: Individual Submission
> Pages: 8
> URL:
> https://www.ietf.org/archive/id/draft-huitema-quic-mpath-option-00.txt
> Status: https://datatracker.ietf.org/doc/draft-huitema-quic-mpath-option/
> Htmlized:
> https://datatracker.ietf.org/doc/html/draft-huitema-quic-mpath-option
> Htmlized: https://tools.ietf.org/html/draft-huitema-quic-mpath-option-00
>
>
> Abstract:
> The initial version of QUIC provides support for path migration. In
> this draft, we argue that there is a very small distance between
> supporting path migration and supporting simulatneous traffic on
> multipath. Instead of using an implicit algorithm to discard unused
> paths, we propose a simple option to allow explicit management of
> active and inactive paths. Once paths are explicitly managed, pretty
> much all other requirements for multipath support can be met by
> appropriate algorithms for scheduling transmissions on specific
> paths. These algorithms can be implemented on the sender side, and
> do not require specific extensions, except maybe a measurement of one
> way delays.
>
>
>
> Please note that it may take a couple of minutes from the time of
> submission
> until the htmlized version and diff are available at tools.ietf.org.
>
> The IETF Secretariat
>
>
>

Reply via email to