CVMFS and HTTP/3, QUIC

Town Square
1

Hello,

Has there been any thought about HTTP/3 and QUIC in the context of CVMFS?
While the reduced number of roundtrips for the handshake process may not be applicable for CVMFS using TCP without TLS, nevertheless there could be potentially significant performance improvements especially for lossy networks, long RTT, as well as improved congestion control algorithms. I’m not sure if head of line blocking affects CVMFS, but HTTP/3 should bring benefits for multiple simultaneous background downloads (bundling?). As always, the actual performance impact may depend on the environment and application and would need to be tested.

About 25% of the web uses HTTP/3 and it has been supported in libcurl for ~ 4 years. However it seems httpd and squid may be late adopters, with codebases from a different era that could make it difficult to add support for new protocols.

https://lists.apache.org/thread/72j2oxvzdy1pvk9zfxd2mljs0kx5vto8

2

Hi Ryan,

indeed these are interesting developments. In the past there have been efforts to try out HTTP/2 and apparently the advantages have not been so compelling that a switch was considered in the end. The problems it adresses are partly mitigated in cvmfs, and the missing support in apache and squid are real showstoppers. That is not to say though that we don’t want to consider it, once the support situation in the caching infrastructure in particular changes this could be fairly easy to add to CVMFS.