Post #1967223
2026-05-04 17:41 UTC
Nice PR landing in picoquic: push the number of supported connection from about 7,000 per core to over 100,000 by fixing the scheduler. Was a simple splay, sufficient in common scenarios, but it is O(logN) and that becomes an issue when the number of connections is larger than usual. Nice fix by Victor Stewart.
https://github.com/private-octopus/picoquic/pull/2095
Replies (1)
-
@huitema@social.secret-wg.org 2026-05-04 17:54
Of course, there are other bottlenecks beside scheduling. The big one is the socket API traversal, which is the limit for "batch" speed trials. But not all applications are sending lots of data. "Presence" style applications are just sending the occasional messages on a connection, and that's why the scheduling bottleneck was an issue.