OxiQUIC 0.1.1 Release

cool-japan released this 04 Jun 13:19

· 2 commits to master since this release

638f90e

[0.1.1] - 2026-06-04

Added

bench_memory_usage benchmark in oxiquic-transport: measures RSS delta per QUIC connection (1 and 10 connections) on Linux (/proc/self/status) and macOS (mach_task_self() task_info); prints a one-time per-connection kilobyte estimate alongside criterion timing data.
bench_h3_memory_profile benchmark in oxiquic-h3: same RSS methodology applied to HTTP/3 connections — reports per-H3-connection heap overhead and measures connection setup rate via criterion.
bench_h3_push_overhead benchmark in oxiquic-h3: documents server push stub latency vs an equivalent client-initiated GET; confirms the push stub path (always NotImplemented in h3 0.0.8) adds no measurable network overhead.
bench_h3_vs_h2_throughput benchmark in oxiquic-h3: sustained throughput comparison of H3 vs H2 at 256 KiB and 1 MiB payload sizes using criterion::Throughput::Bytes to report bytes/s; exercises flow-control and congestion-window paths.

Fixed

oxiquic-transport driven connection loop: io::ErrorKind::ConnectionRefused (ICMP port-unreachable, sent when the peer socket closes before a CONNECTION_CLOSE frame) is now treated as non-fatal — the loop continues instead of breaking, letting the QUIC loss-detection timer handle recovery per RFC 9000.
oxiquic-h3 h3_get_roundtrip integration test: server now calls h3_conn.shutdown(0) after serving the response, so the client receives CONNECTION_CLOSE and the driver task exits cleanly instead of racing against the QUIC idle-timeout.

Full Changelog: v0.1.0...v0.1.1

Assets 2