73 views
 owned this note
# MIR QUIC Breakout (2023-11-23) ## Problems of the Protocol Spec - amplification limits difficult to work with - Breaks the paradigm of transport:port - multiplexing is above the port-level - breaks flow identification - Connection-IDs too short to encode information for stateless processing at middleboxes etc. - multiple extensions being in use at the same time: it's not defined how they influence each other. Extensions should not be required in the future to be RFC-9000 compliant to allow QUIC to be used in constrained environments. - it's not specified if the max streams value can be set to 0? could QUIC streams just be an extension / could QUIC simply provide an encrypted connection? - did QUIC specify too many things which should be part of HTTP/3? Like streams. ## Problems of QUIC Implementations ### User-space vs kernel - QUIC is implemented in user-space and (currently) not in the kernel - Kernel implementation of Crypto?? - adoption issue? - API - relation to TAPS (-> Michael?)? - Apple uses some kind of TAPS with QUIC in the Kernel ### Congestion Control - congestion control implementations: every QUIC library needs to implement them again - which is the "right" way to implement a specific algorithm? - fairness between different implementations ### Interop - currently compatibility and performance problems ## Problems of QUIC Deployment - Default UDP-socket buffer size in Linux - How much UDP-blocking is around? - Load balancing with initial address update/Load balancing extension ## Research Questions, Objectives for Working on QUIC - Measuring the QUIC evolution - Kernel space implementation (without crypto) - Multipath QUIC vers. Multipath TCP (aka scheduling, performance, RTT calculation, loss recovery ...) ### Multicast QUIC - Scalability and key management [1,2] - (Multipath sessions including multicast channels) - Multicast and use control channel for selecitve retransmissions - Some related work - [1] L. Navarre, O. Pereira, and O. Bonaventure, “MCQUIC: Multicast and unicast in a single transport protocol.” 2023. - [2] M. Franke, J. Holland, and S. Schmid, “MCQUIC – A Multicast Extension for QUIC.” 2023. - IETF presentation from Franke on `draft-jholland-quic-multicast`: https://datatracker.ietf.org/meeting/118/materials/slides-118-mboned-mcquic-00