Lucas,
Many thanks for these resources. I'm slowly reading and digesting.
Quentin's Thesis I think has some good discussion about consideration
factors for an MP-QUIC abstract design, such as consideration of path
validation etc. I don't know if uniflow is the correct design but it
definitely helps to understand the motivation for it.
MP-QUIC appears to offer some things that conventional QUIC might not be
able to do*. But it, to me, also presents some problems that don't exist
today. As an individual commenter, not having an in-band signal for
uniflow prioritisation gives me cause for concern. If I were to deploy
an MP-QUIC general-purpose server, I don't have much confidence that I
could meet the use cases that clients want.
Prioritisation can be easily added to MPQUIC. It was not included in the
current draft because we focused on the basic principles and did not
want to add new features before the start of technical discussions on
the mailing list. There are different ways to do prioritisation,
starting from the one-bit solution used by MPTCP to other scenarios like
putting an integer weigth on uniflows or limiting the bandwidth on a per
uniflow basis, ...
Olivier
