Original version
Computer Networks. 2020, 173:107211, DOI: https://doi.org/10.1016/j.comnet.2020.107211
Abstract
Network APIs are moving towards protocol agility, where applications express their needs but not a static protocol binding, and it is up to the layer below the API to choose a suitable protocol. The IETF Transport Services (TAPS) Working Group is standardizing a protocol-independent transport API and offering guidance to implementers. Apple’s recent “Network.framework” is specifically designed to allow such late and dynamic binding of protocols. When the network stack autonomously chooses and configures a protocol, it must first test which protocols are locally available and which work end-to-end (“protocol racing”). For this, it is important to know the set of available options, and which protocols should be tried first: Does it make sense to offer unchecked payload delivery, as with UDP-Lite? Is a UDP-based protocol like QUIC always a better choice, or should native SCTP be tried? This paper develops answers to such questions via (i) a NAT study in a local testbed, (ii) bidirectional Internet tests, (iii) a large scale Internet measurement campaign. The examined protocols are: SCTP, DCCP, UDP-Lite, UDP with a zero checksum and three different UDP encapsulations.