libsrt Comparison
Date: 2026-03-27 Last updated: 2026-08-11
Note on versions: bilbycast-srt advertises wire compatibility with libsrt v1.5.5 (0x010505), and this comparison is written against that release. Upstream has since published v1.5.6, a security release fixing two high-severity CVEs that affect every version up to and including 1.5.5.
The bilbycast-libsrt-rs wrapper — which is what bilbycast-edge actually builds against — vendors v1.5.6, so shipped edge binaries are not exposed to those CVEs. If you build against a system libsrt, use 1.5.6 or later.
Overview
Section titled “Overview”| bilbycast-srt | libsrt v1.5.5-rc | |
|---|---|---|
| Language | Rust (no external C deps) | C++ with C API |
| Architecture | 3 crates: srt-protocol (no I/O), srt-transport (tokio), srt-ffi (WIP) |
Monolithic C++ library |
| Async model | Tokio async/await | Epoll + threads |
| Crypto deps | RustCrypto | OpenSSL-EVP (default), mbedTLS, GnuTLS, Botan |
| Version advertised | 0x010505 (1.5.5) |
0x010505 (1.5.5) |
Feature Parity
Section titled “Feature Parity”| Feature | bilbycast-srt | libsrt v1.5.5 | Notes |
|---|---|---|---|
| HSv5 Handshake | Yes | Yes | Induction + Conclusion, extension blocks |
| Caller mode | Yes | Yes | |
| Listener mode | Yes | Yes | |
| Rendezvous mode | Yes | Yes | |
| Live mode (TSBPD) | Yes | Yes | Both default 120ms latency |
| File mode (AIMD) | Yes | Yes | |
| AES-CTR encryption | Yes | Yes | |
| AES-GCM (AEAD) | Production | Preview | bilbycast-srt ships GCM as first-class; libsrt still behind ENABLE_AEAD_API_PREVIEW build flag, AEAD epic #2336 still open |
| Key sizes 128/192/256 | Yes | Yes | |
| PBKDF2 key derivation | Yes | Yes | HMAC-SHA1, 2048 iterations |
| AES Key Wrap (RFC 3394) | Yes | Yes | |
| Key rotation (even/odd) | Yes | Yes | 16M pkt refresh, 4096 pkt pre-announce |
| Enforced encryption | Yes | Yes | |
| ARQ (NAK retransmit) | Yes | Yes | |
| FEC (row-only) | Yes | Yes | Config: "fec,cols:10,rows:1". Negotiated via SRT_CMD_FILTER (ext type 7) in handshake. |
| FEC (staircase/2D) | Yes | Yes | Config: "fec,cols:10,rows:5,layout:staircase". Staircase column offsets match libsrt. 2D cascade recovery. |
| FEC ARQ modes | Yes | Yes | arq:always (parallel), arq:onreq (FEC-first, default), arq:never (FEC-only). NAK suppression matches libsrt. |
| Too-Late Packet Drop | Yes | Yes | |
| Stream ID (send + receive) | Yes | Yes | Up to 512 chars. Caller sends SRT_CMD_SID (ext type 5) in CONCLUSION; listener parses and stores. Structured #!::key=value format parsed via StreamIdInfo. |
| Stream ID on accepted socket | Yes | Yes | bilbycast-srt: socket.stream_id() getter; libsrt: srt_getsockopt(SRTO_STREAMID) |
| Drift tracking | Yes | Yes | bilbycast-srt: 1000-sample window |
| Statistics (80+ counters) | Yes | Yes | Includes FEC stats (recovered/lost/overhead), ACK/NAK, flow control, buffer state, TSBPD delays, reorder metrics |
| Epoll multiplexing | Yes | Yes | |
| Bidirectional data | Yes | Yes | |
| NAK report | Yes | Yes | |
| Loss max TTL | Yes | Yes | Reorder tolerance |
| Access Control callbacks | Yes | Yes | bilbycast-srt: AccessControl trait + access_control_fn() closure API with HandshakeInfo (peer addr, stream ID, encryption state); libsrt: srt_listen_callback() with SRTSOCKET handle |
| Retransmit bandwidth cap | Yes | Yes | Both use Token Bucket algorithm; bilbycast-srt: max_rexmit_bw config (SRTO_MAXREXMITBW); libsrt: CShaper class |
| Rejection with reason codes | Yes | Yes | bilbycast-srt sends HandshakeType::Failure(RejectReason) on reject; 18 reason codes matching libsrt |
| Socket Groups / Bonding | No | Yes | Broadcast + Main/Backup (Balancing still WIP) |
| Auto startup/cleanup | N/A | Yes | New in v1.5.5 — automatic global init (Rust doesn’t need this) |
| Windows ARM64 | Untested | Yes | New in v1.5.5 |
| HarmonyOS (OHOS) | No | Yes | New in v1.5.5 |
| C FFI | WIP | Native | libsrt’s primary interface |
Access Control — Implementation Comparison
Section titled “Access Control — Implementation Comparison”| Aspect | bilbycast-srt | libsrt v1.5.5 |
|---|---|---|
| API style | AccessControl trait or closure via access_control_fn() |
C callback via srt_listen_callback() |
| Info provided | HandshakeInfo { peer_addr, stream_id, is_encrypted, peer_socket_id, peer_version } |
SRTSOCKET handle (query any socket option) |
| Rejection | Returns Err(RejectReason) — 18 standard codes |
Returns -1 with srt_setrejectreason() — same codes |
| Stream ID format | Sends/parses SRT_CMD_SID extension (type 5) in CONCLUSION. StreamIdInfo parses #!::key=value format (keys: r, m, s, t, u, h). |
Same wire format |
| Per-connection passphrase | Not yet (callback sees is_encrypted but can’t override) |
Yes — callback can call srt_setsockopt(SRTO_PASSPHRASE) |
| Stored on accepted socket | socket.stream_id() getter |
srt_getsockopt(SRTO_STREAMID) |
Token Bucket Shaper — Implementation Comparison
Section titled “Token Bucket Shaper — Implementation Comparison”| Aspect | bilbycast-srt | libsrt v1.5.5 |
|---|---|---|
| Algorithm | Classic Token Bucket | Token Bucket (CShaper class) |
| Config | max_rexmit_bw: i64 in SrtConfig |
SRTO_MAXREXMITBW socket option |
| Rate values | -1 = unlimited (default), 0 = disable retransmit, > 0 = bytes/sec |
-1 = unlimited, 0 = disable, > 0 = bytes/sec |
| Burst sizing | max(10ms of bandwidth, 2 * MSS) |
Similar heuristic in CShaper |
| Integration point | send_retransmissions() — checks bucket per packet, defers rate-limited packets |
CSndQueue::worker — similar per-packet gating |
| Builder API | .max_rexmit_bw(bytes_per_sec) on socket and listener builders |
srt_setsockopt(SRTO_MAXREXMITBW) |
What Changed from v1.5.4 to v1.5.5
Section titled “What Changed from v1.5.4 to v1.5.5”| v1.5.5 Change | Impact on comparison |
|---|---|
| Token Bucket for MAXREXMITBW | Parity — bilbycast-srt now has its own TokenBucket shaper for max_rexmit_bw |
Thread safety fixes (lock-free m_bListening, shared mutex, fork safety, strerror reentrancy) |
Non-issue for bilbycast-srt — Rust’s ownership model prevents these classes of bugs |
| Cookie contest restored from v1.4.5 | bilbycast-srt should verify its cookie contest logic matches the restored behavior for interop |
| Blocking srt_connect error codes fixed | N/A — bilbycast-srt is async, no blocking API |
| Buffer overflow fix in handshake group data | N/A — bilbycast-srt doesn’t implement groups; Rust would catch this at bounds check anyway |
| Late-rejection for mismatched packet filter | Parity — bilbycast-srt rejects with RejectReason::Filter when FEC parameters conflict |
| CMake LIBSRT_ prefix | N/A — Cargo workspace, no CMake |
| Windows ARM64 + HarmonyOS | Platform gap — bilbycast-srt compiles on any Rust target but hasn’t been tested on these |
Where bilbycast-srt Is Ahead
Section titled “Where bilbycast-srt Is Ahead”- AES-GCM is production-ready — libsrt v1.5.5 still gates GCM behind a preview build flag with open issues (TSBPD required, listener can’t force GCM mode, epic #2336 incomplete).
- Memory safety by construction — Many v1.5.5 fixes (buffer overflow, data races, lock-order inversions, reentrancy bugs, use-after-free patterns) are structurally impossible in Rust.
- Clean protocol/transport separation —
srt-protocolhas zero I/O dependencies, embeddable in any runtime (WASM, no_std, custom event loops). libsrt tightly couples protocol with threading. - Tokio-native async — Natural fit in Rust async ecosystems. No thread pool management.
- Zero system dependencies — No OpenSSL, no pkg-config, no C toolchain. Single
cargo build. - Ergonomic access control API — Rust trait + closure API is more composable than C callback.
HandshakeInfostruct provides typed fields rather than requiring socket option queries.
Where libsrt v1.5.5 Is Ahead
Section titled “Where libsrt v1.5.5 Is Ahead”- Socket Groups / Bonding — Broadcast and Main/Backup for hitless failover. The only remaining major feature gap. (Balancing mode still WIP even in libsrt.)
- Per-connection passphrase override — libsrt’s
srt_listen_callbackcan setSRTO_PASSPHRASEper connection; bilbycast-srt’s access control can accept/reject but not override the passphrase dynamically. - C FFI maturity — Used by FFmpeg, OBS, GStreamer, VLC. bilbycast-srt’s FFI is scaffolding.
- Platform breadth — Now includes Windows ARM64 and HarmonyOS. bilbycast-srt is untested on mobile/embedded.
- Ecosystem adoption — De facto industry standard with broad tooling support.
Interop Considerations
Section titled “Interop Considerations”Since bilbycast-srt advertises version 0x010505, it should verify:
- Cookie contest logic matches the restored v1.4.5 method (changed in v1.5.5)
Packet filter late-rejection handling— Done: bilbycast-srt rejects withRejectReason::Filterwhen FEC parameters conflict during negotiation- AES-GCM 12-byte IV (changed in v1.5.4, carried into v1.5.5)
- Stream ID extension interop — bilbycast-srt now sends and parses
SRT_CMD_SID(ext type 5) in CONCLUSION; verify round-trip with libsrt callers and listeners
Summary
Section titled “Summary”bilbycast-srt now matches libsrt v1.5.5 on access control, retransmission bandwidth shaping (token bucket), Stream ID sending and parsing, structured Stream ID format (#!::key=value), and FEC (row-only, staircase/2D, ARQ integration, handshake negotiation). The only remaining major feature gap is bonding/socket groups. bilbycast-srt leads on AES-GCM maturity, memory safety, architectural cleanliness, and API ergonomics.
For bilbycast’s use case (media transport gateway with its own relay infrastructure for redundancy), the bonding gap is mitigated by bilbycast-edge’s own hitless redundancy and tunnel failover mechanisms.
Feature coverage: ~95% of libsrt v1.5.5
Section titled “Feature coverage: ~95% of libsrt v1.5.5”| Category | Coverage |
|---|---|
| Core protocol (handshake, ARQ, TSBPD, timers) | 100% |
| Encryption (AES-CTR, AES-GCM, key rotation) | 100% (GCM ahead) |
| FEC (row, staircase, ARQ integration) | 100% |
| Congestion control (Live, File) | 100% |
| Access control (Stream ID, accept/reject) | ~90% (missing per-connection passphrase override) |
| Retransmit shaping (Token Bucket) | 100% |
| Connection modes (caller, listener, rendezvous) | 100% |
| Bonding / socket groups | 0% |
| C FFI | ~10% (scaffolding) |