mediasoup vs HOMER

The Rust crate ships small, frequent releases carrying one or two commits each. A single thread runs through the current window: subchannel support over pipe transport, which started with encoding subchannels into SCTP messages in 0.23.0, continued with pipe DataConsumers handling them in 0.24.2, and reached the API surface in 0.25.0 with an ignored_subchannel argument on DirectDataProducer.send(). Alongside it, 0.24.0 was a security batch — constant-time comparison for SCTP cookie and STUN message-integrity MACs, an out-of-bounds write in RTP dependency descriptor parsing, an SCTP integer overflow — with an outside researcher credited on all three.

The work is squarely on multi-worker and multi-host routing: subchannels let a single data stream be selectively delivered, and making that work across pipe transport is what lets it survive the hop between workers. That is scaling infrastructure for SFU operators running mediasoup across processes or machines rather than a feature end users see. The parallel security hardening suggests the library is getting outside scrutiny, and the SVC enablement for VP8 and H264 in 0.22.11 rounds out the same picture — media quality and routing capability for large deployments.

Expect the subchannel work to continue closing gaps in the pipe transport path, since each release in this sequence has handled one more place subchannels were dropped. The one-or-two-commit release habit means version numbers will keep climbing faster than the scope of any individual change.