欢迎致辞及主题演讲——高层次、不局限于特定目标的运输方案.pdf

1、High-Level and Target-Agnostic Transport ProgramsMina Tahmasbi ArashlooUniversity of WaterlooP4 Workshop 2025No“one-size-fits-all”transport protocolApplicationNetwork(best-effort packet-based)TransportApplication-layer(arbitrary-sized)data(Paced)Data segmentswhich data segment to send and when such

2、that Data is reliably delivered to the receiver as fast as possible w/o overwhelming the network and receiverNo“one-size-fits-all”transport protocolApplicationNetwork(best-effort packet-based)TransportApplication-layer(arbitrary-sized)data(Paced)Data segmentswhich data segment to send and when such

3、that Data is reliably delivered to the receiver as fast as possible w/o overwhelming the network and receiverDepends on Network characteristics Wide area?Data center?Applications Traffic patterns:small flows?Bursty?Requirements:low latency?High throughput?No“one-size-fits-all”transport protocolAppli

4、cationNetwork(best-effort packet-based)TransportApplication-layer(arbitrary-sized)data(Paced)Data segmentsTCP and its many many variantsQUICNDPHomaRoCEv2The transport protocol development cycle todayEnsure it works as intendedPick the“right”protocol/featuresOptimizeImplement on your“target”No high-l

5、evel specification with well-defined semantics Natural language documents ambiguity Existing implementations low-level target-specific codeNo high-level specification with well-defined semantics Intended behavior is not always clear Pick and choose scenarios to test No automated high-coverage analys

6、is and testingHave to grapple with low-level protocol-independent issues I/O,memory management,optimized data structures,The transport protocol development cycle todayEnsure it works as intendedPick the“right”protocol/featuresOptimizeImplement on your“target”No high-level specification with well-def