Path diversity, ARQ and active-active vs active-standby for resilient live contribution

How it works

Dual‑path seamless protection (ST 2022‑7)

SMPTE ST 2022‑7 defines a mechanism in which two independent RTP streams — commonly called A/B or red/blue legs — carry identical RTP packets over distinct physical or logical networks. The receiver performs packet-level reconstruction by selecting the first-arriving copy of each RTP sequence number and outputs a single, ordered RTP stream. When one path drops packets, the receiver continues delivering without frame loss provided the alternate path contains the corresponding packets and PTP timing remains within buffer bounds.

Hitless switching requires synchronised clocks (PTP) and buffers sized to absorb the worst-case differential delay between paths. Typical buffer headroom is selected from measured path skew, which can range from tens to hundreds of milliseconds depending on WAN diversity. The formal definitions and requirements are specified in SMPTE ST 2022‑7. In practice, this mechanism is the gold standard for uncompressed, PTP-synchronised studio links where any perceptible interruption is unacceptable.

Path bonding / multi‑path aggregation (per‑flow or per‑packet)

Path bonding can operate in two fundamentally different modes, each with distinct implications for timing, reordering, and resilience.

Per-flow bonding pins flows — identified by RTP SSRC or 5-tuple — to a single path for the life of a session. This minimises packet reordering and is predictable for session-oriented transports, but it does not average bandwidth or packet loss across links. It is the preferred mode for tightly time-synchronised environments such as SMPTE ST 2110 where session persistence and strict timing are paramount.

Per-packet bonding spreads individual packets across multiple interfaces to aggregate capacity and statistically reduce the impact of any single path’s loss. The trade-off is increased out-of-order delivery, which demands resequencing buffers at the receiver — typically adding tens to hundreds of milliseconds of latency. This mode is best suited for highly compressed WAN contribution (HEVC/AVC) over unmanaged networks where throughput aggregation is the primary goal and the receiver supports adequate jitter buffers.

A practical rule: use per-flow bonding for session persistence and tight timing, and reserve per-packet modes for best-effort compressed transport where resequencing latency is tolerable.

Packet-recovery building blocks: ARQ, FEC and hybrids

Three recovery strategies underpin every redundancy design, each with a distinct latency/bandwidth profile. ARQ (selective retransmission) resends only what was lost but costs at least one round-trip per recovery cycle, which rules it out for sub-frame latency budgets. FEC (SMPTE ST 2022-1/5) transmits parity proactively — typically 10-20% overhead, configurable higher for severe burst profiles — recovering loss without any RTT dependency at the cost of bandwidth and FEC-block decoding latency. Hybrid schemes pair lightweight FEC for routine loss with ARQ for the rare bursts that exceed parity capacity, tuned from historical loss and burst metrics.

The retransmission mechanics, receive-buffer sizing formula, and the full ARQ-versus-FEC decision logic are covered in depth in SRT Packet Loss in Live Video Contribution; this article treats them as building blocks and concentrates on path-level design — how many paths, in what operating mode, and with what switchover behaviour.

Active‑active vs active‑standby operational behaviour

The choice between active-active and active-standby operation has cascading effects on availability, bandwidth cost, and operational complexity.

In an active-active configuration, all paths carry traffic simultaneously. The receiver merges or decodes from whichever path delivers each packet first. When path failure rates are independent, outage probability improves multiplicatively — two paths each at 99.9% availability yield a combined availability well above 99.999%. The cost is higher aggregate egress bandwidth, increased synchronisation requirements, and more complex resequencing logic. Active-active pairs naturally with FEC or hitless switching (ST 2022‑7).

In an active-standby configuration, only the primary path carries traffic; the standby path is idle or carries low-rate health probes until failover is triggered. This is simpler to implement and reduces bandwidth cost, but failover introduces detect-and-switch time and possible micro-outages unless hitless switching is supported at the RTP level.

When defining failover behaviour, specify two metrics: allowable switchover time (RTO) and acceptable transient loss (RPO). For live broadcast, RTO targets are commonly sub-second to a few seconds depending on programme tolerance and contractual obligations.

Why it matters in broadcast

Brief outages or lip-sync errors during high-value content — sport, news, live events — directly affect revenue, regulatory compliance, and contractual delivery obligations. Redundancy choices translate into quantifiable business risks: the cost per minute of outage for a tier-one sports feed can dwarf the cost of a second diverse path.

Remote production and interfacility transport often require sub-frame or low-single-digit-millisecond added latency for multi-camera switching and PTP synchronisation. Some redundancy mechanisms, notably ARQ, are incompatible with the tightest latency budgets, making the choice of protection strategy inseparable from the production workflow design.

Centralised IP models enable unified failover and observability across many feeds, but the chosen redundancy approach must scale without multiplying operational overhead or energy draw across dozens or hundreds of simultaneous contribution streams. Using established standards — ST 2110, ST 2022‑1/5/7, SRT — simplifies multi-vendor workflows and reduces integration risk, a critical consideration as broadcasters modernise from satellite-centric to IP-based architectures.

Technical specifications / trade‑offs

Comparison of redundancy alternatives

Alternative	Typical use case	Bandwidth cost	Latency impact	Protection type	Best for
ST 2022‑7 dual‑path (hitless)	Uncompressed ST 2110 / professional IP links	100% duplicate (A+B)	Sub-frame to low ms (with PTP sync and buffering)	Seamless hitless switching at RTP level	Uncompressed, PTP-synchronised studio links
Per‑flow bonding (session pinning)	Compressed contribution requiring session persistence	Single path per session (no duplication)	Minimal (normal jitter buffer only)	None inherent; combine with standby or FEC	Remote production with session affinity and tight sync
Per‑packet bonding / multi‑path aggregate	WAN bonding for throughput or resilience	Aggregate of link capacities; no duplication unless configured	10s–100s ms (resequencing buffer)	Statistical resilience; reduced outage probability	HEVC/AVC over unmanaged networks
ST 2022‑1 / FEC (1D/2D parity)	Lossy WAN with burst losses	Parity overhead 10-20% (configurable higher)	Tens to hundreds of ms (FEC block window)	Burst recovery without RTT dependence	Predictable burst loss, constrained RTT
ARQ (retransmit)	Managed networks with low RTT	Low (retransmits only)	≥ 1 RTT per recovery (hundreds of ms on WAN)	Exact recovery of missing packets	Non-sub-frame latency; quality over latency

Sizing and availability estimation

A conservative approach to sizing redundancy starts with measurable inputs: required availability target A (e.g. 99.99%), single-path availability p (e.g. ISP SLA of 99.9%), an independence factor i (e.g. 0.9 for shared geography), the number of simultaneous feeds N, and cost per Mbps C.

For an active-standby two-path configuration, approximate combined availability is:

A ≈ 1 − (1 − p) × (1 − p × i)

With p = 0.999 and i = 0.9, this yields A ≈ 1 − (0.001) × (1 − 0.0009) ≈ 0.9999991, or roughly 99.99991%. The geometric improvement with independent paths is dramatic — but reduce i if paths share physical risk (last-mile fibre, regional exchange).

Practical guidance: require at least two diverse physical carriers for broadcast-grade contribution. Add a third path for higher SLA targets or to permit maintenance without risking service. Use measured historical loss and mean time to repair (MTTR) to tune buffer sizes and FEC parity ratios.

Operational controls and runbook essentials

Redundancy without observability and automated controls creates hidden operational risk. The following controls should be embedded in any resilient contribution architecture:

• Health metrics: Continuous monitoring of packet loss (1s, 10s, 60s windows), jitter, RTT, PTP offset, and path differential delay. Alert thresholds should map to automated actions.

• Automatic switchover policy: Define detection thresholds (loss rate, jitter spike, or complete path failure), minimum dwell time to avoid flapping, and failback policy (immediate vs. manual confirmation).

• Test regimes: Monthly path-fail tests for critical feeds, quarterly for lower-criticality feeds. Include simulated burst loss with FEC/ARQ parameter sweeps, and rehearse failback with exact timings recorded. Track mean time to failover (MTTFo) and MTTR as operational KPIs.

• Observability infrastructure: Per-stream telemetry, centralised dashboards, and synthetic traffic probes to pre-validate alternate paths before they are needed.

Related approaches or standards

Several standards and protocols form the ecosystem around resilient IP contribution:

• SMPTE ST 2022‑7 — Seamless protection switching of RTP datagrams, providing hitless dual-path reconstruction.

• SMPTE ST 2022‑1 and ST 2022‑5 — Forward error correction mechanisms for real-time media transport using 1D and 2D parity.

• SMPTE ST 2110 — The suite for professional media over IP, defining timing and transport conventions for audio, video, and ancillary data.

• SRT (Secure Reliable Transport) — an ARQ-style retransmit protocol widely used for encrypted, retransmit-capable contribution over unmanaged networks; its recovery behaviour is examined in SRT Packet Loss in Live Video Contribution.

• NMOS IS‑04/05 — Discovery and connection management specifications for ST 2110 environments, enabling automated resource registration and path management in redundant topologies.

How Appear addresses this

Appear applies protection at the transport layer, in hardware on the X Platform and in software with VX.

ST 2022-7 is the mechanism Appear leans on for active-active operation. Both platforms reconstruct from dual paths at the RTP layer, and on the X Platform seamless protection runs over SRT as well as plain IP, using a single seamless buffer spanning both paths rather than a separately protected buffer per path. X10 and X20 carry it at density, X5 covers the venue end, and VX provides the same seamless merge plus A/B failover on commodity Linux hardware, private-cloud VMs, or public-cloud instances.

FEC is scoped more narrowly than the comparison above might suggest. ST 2022-1 FEC — column-only or column-and-row, with configurable matrix dimensions — is available on the IP gateway path for TSoIP. Appear’s SRT is an in-house implementation that recovers by ARQ and carries no FEC filter, so on an SRT leg the receive buffer rather than a parity ratio is the parameter to size. On X5, ST 2022-1 FEC is available on output only.

Bonding on the X Platform is active-backup. L2 bonding with the SRT image provides path failover in floating or reverting mode, not per-packet aggregation across links. In the terms of this article that is an active-standby control rather than a per-packet bonding one — a distinction that matters when a design assumes aggregated capacity.

FAQ

How many redundant paths do I need for a single broadcast feed?
A minimum of two physically and logically diverse paths provides basic broadcast-grade resilience. Add a third path when you require maintenance windows without service risk, or when availability targets demand it after calculating path independence and the business cost of downtime.

When should I choose active-active over active-standby?
Choose active-active when maximum availability is required and you can support duplicate stream bandwidth plus receiver resequencing. Choose active-standby when bandwidth cost or receiver simplicity is paramount and short switchover times are acceptable.

Does ST 2022‑7 eliminate all perceptible interruption?
ST 2022‑7 provides hitless switching if receiver buffers and clock synchronisation are correctly configured. It cannot mask correlated failures that affect both paths simultaneously or losses that exceed the buffering headroom.

How do I dimension FEC parity and block sizes?
Base parity ratio and block sizes on measured loss rate and burst length. Increase parity for higher burst probability; increase block size for better burst recovery, but expect proportionally larger decoding latency.

Can ARQ be used for sub-frame latency workflows?
Generally no. ARQ introduces at least one RTT of recovery delay and is unsuitable for sub-frame or ultra-low latency use cases. Use FEC or hitless dual-path switching for those constraints.

How important is physical path diversity versus multiple ISPs?
Very important. True independence requires separate physical routes and ideally different backbone providers. Multiple ISPs that share last-mile infrastructure or regional exchanges can still share failure domains, significantly reducing the independence factor.

What telemetry should I collect to drive automated failover?
Per-path packet loss, jitter, RTT, PTP offset, and path differential delay, plus hardware/port state and ISP-level alerts. Use short windows (1–10s) for fail detection and longer windows (60s+) for trend analysis and capacity planning.

How do I avoid split-brain or flapping during automated failover?
Use hysteresis (dwell time), quorum-style health checks, and assisted signalling via a control channel before switching. Implement conservative thresholds and exponential backoff for repeated failover events.

Is per-packet bonding compatible with ST 2110 uncompressed video?
Per-packet bonding usually causes resequencing that can break the strict timing requirements of uncompressed ST 2110, unless the receiver and network are engineered with sufficient resequencing buffers and timing management — which adds complexity and latency.

What business metrics should influence redundancy spend?
Use cost per hour of downtime, contractual SLA penalties, viewer and revenue sensitivity, and brand risk to convert availability targets into a budget for additional paths, bandwidth, and monitoring infrastructure.

How often should disaster recovery failover tests run?
Monthly for critical feeds, quarterly for lower-criticality feeds. Periodically perform blind failover tests (operator absent) and record RTO/RPO to validate runbooks against actual performance.

How does encryption interact with redundancy techniques?
Encryption is orthogonal to redundancy. Apply end-to-end encryption per path (e.g. AES-256) and ensure key management scales across redundant paths. Encrypted streams must still include consistent RTP headers and timing information to allow hitless reconstruction at the receiver.