Skip to content

Ask for 8 MB UDP socket buffers on the raw-QUIC client on Linux - #67

Merged
rwl4 merged 1 commit into
mainfrom
feat/udp-receive-buffer
Nov 16, 2026
Merged

rwl4 merged 1 commit into
mainfrom
feat/udp-receive-buffer

Conversation

@gwendalsimon

@gwendalsimon gwendalsimon commented Nov 12, 2026 •

Copy link
Copy Markdown
Contributor

On Linux, the raw-QUIC client of moq_endpoint now asks picoquic for 8 MB UDP socket buffers (socket_buffer_bytes). The kernel default of about 208 KB overflowed on a receiver: RcvbufErrors grew by about 2 per second per socket. In a one-hour A/B test on 10 November 2026, a receiver with 8 MB buffers (set through rmem_default) dropped 0 packets, against 7,233 with 208 KB.

Linux silently caps the request at twice net.core.rmem_max. A stock host therefore gets about 416 KB. The full 8 MB needs the operator to raise that limit. Callers need no change. Other systems keep their default, because macOS refuses a size above kern.ipc.maxsockbuf and the socket would then not open. A new test checks the value the adapter receives.


This change is Reviewable

The default UDP receive buffer (about 208 KB) overflowed on a receiver
at media rates. The raw-QUIC client of the service endpoint now asks
picoquic for 8 MB socket buffers on Linux, which caps the request at
net.core.rmem_max without failing. Other systems keep their default,
because macOS refuses a size above kern.ipc.maxsockbuf.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@rwl4
rwl4 merged commit 64f5c01 into main Nov 16, 2026
31 of 37 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants