Technical Article
Choosing an SRT Platform for News Broadcast Contribution and Ingest
Secure Reliable Transport (SRT) platforms are systems—software, appliances, or cloud services—that implement the SRT protocol to deliver encrypted, packet-repaired, and latency-tuned live video over unmanaged IP networks. For newsroom ingest, the right platform is not simply the one with the highest headline specification; it is the one whose capacity, latency tuning range, redundancy architecture, and operational interfaces align with real newsroom constraints. Those constraints include the number of concurrent ENG and bureau feeds during peak events, breaking-news latency targets that keep reporter-to-anchor interactions feeling immediate, automated failover behaviour when a high-priority feed must pre-empt others, and integration with newsroom automation and IP production control. This article provides a technical framework for evaluating and comparing SRT platforms against those requirements.
How it works
SRT platforms implement three functional layers that matter for newsrooms: transport reliability, latency control, and operational scale. Each layer can be realised in dedicated hardware, software running on commodity servers, managed cloud services, or hybrid combinations. Understanding the mechanics of each layer is essential before comparing products.
Transport reliability and packet recovery
SRT can encrypt media payloads with AES (128-, 192-, or 256-bit keys, depending on the implementation) and applies an Automatic Repeat reQuest (ARQ) mechanism with selectable retransmission window sizes. When a receiver detects a missing packet it requests retransmission from the sender within the configured window. This dramatically reduces the quality impact of packet loss compared with plain UDP streaming, but it introduces a trade-off: wider retransmission windows recover more packets at the cost of additional delay, while narrower windows keep latency low but tolerate less loss before visible or audible artefacts appear.
Production-grade platforms typically layer additional protection on top of SRT’s native ARQ:
-
Forward error correction (FEC) adds redundant data so that a proportion of lost packets can be reconstructed without waiting for retransmission.
-
Dual-path or ST 2022-7-style redundant streams send identical copies of each flow over diverse network paths, enabling hitless protection for critical news feeds.
The combination of ARQ, FEC, and path diversity gives engineers a tuneable reliability stack, but each mechanism consumes bandwidth and processing headroom that must be factored into platform sizing.
Latency control and buffering
SRT exposes a configured latency buffer, typically set in milliseconds. It does not adapt on its own: sender and receiver each set a value, the connection uses the larger, and the usual guidance sets it as a multiple of the measured round-trip time (RTT), commonly three times RTT at low loss and four times or more as loss rises. The buffer absorbs jitter and provides time for ARQ retransmissions to complete before the playout deadline. Lower buffers reduce end-to-end delay but increase sensitivity to jitter spikes and burst loss.
For breaking-news ingest where sub-second or near-sub-second latency is essential, engineers configure small ARQ windows, minimal or no FEC, integer-frame-aligned playout, and apply network QoS where the path supports it. For reliability-focused feeds—such as long-haul bureau links traversing multiple ISP hops—a larger buffer (often hundreds of milliseconds) and stronger FEC are standard practice. The key engineering decision is matching the latency-reliability trade-off to each feed’s editorial role: a live two-way demands minimum delay, while a pre-recorded package feed tolerates more buffering in exchange for resilience.
Capacity and concurrency
An SRT platform must be sized for two related but distinct metrics: aggregate throughput (measured in Gb/s) and concurrent connection count (the number of simultaneous SRT sessions). Hardware appliances typically publish per-chassis SRT throughput ceilings and simultaneous-connection limits. Cloud and software platforms scale horizontally by adding instances, but each instance is constrained by CPU cycles and packets-per-second (pps) limits.
In practice, the load a platform must carry equals the sum of per-feed bitrates across all active sessions. Engineers should provision 20–50 % headroom above that sum to accommodate spikes during codec changes, simultaneous multi-bitrate outputs, or sudden influxes of feeds during breaking events. Undersizing either throughput or connection count leads to CPU saturation, dropped connections, or forced re-encodes—any of which can take a live feed off air. Retransmissions add to the load on lossy paths, so size each SRT link above the stream bitrate as well.
Operational control and automation
Newsrooms require programmatic control over their ingest estate. Key integration points include:
-
APIs for session creation and teardown, enabling automation systems to spin up or release feeds without manual intervention.
-
Metadata and routing integration via MOS, NMOS (IS-04/IS-05), or REST interfaces so that feed identity, technical parameters, and routing decisions propagate through newsroom automation.
-
Real-time monitoring of RTT, packet loss, jitter, and buffer occupancy with alerting thresholds and automated cutover rules.
Hot-replacement workflows for breaking items—where a high-priority feed must override a lower-priority one—demand orchestration support: API-triggered routing changes, priority pre-emption logic, or automated tie-breaking rules in the ingest gateway. Without these, operational staff must manually reconfigure routing under time pressure, which is a reliability risk during the moments that matter most.
Why it matters in broadcast
News operations commonly ingest dozens of concurrent ENG and mobile feeds into a single ingest estate during major events. Insufficient connection capacity or raw packet-processing limits cause CPU saturation, dropped connections, or forced re-encodes—failures that directly affect on-air output.
Breaking-news latency tolerance is an editorial concern as much as a technical one. When a reporter and anchor interact live, perceptible delay undermines the conversational feel and can cause talk-over errors. Balancing latency against robustness directly affects editorial agility: too much buffer and the interaction feels sluggish; too little and the feed tears on a lossy path.
Rapid cutover and reliability are non-negotiable. When an unfolding event demands that a high-priority feed override others, the ingest system must support deterministic, automated failover and priority pre-emption without human-intensive reconfiguration. Manual switchovers measured in minutes are incompatible with the pace of live news.
Finally, operational cost and predictability shape long-term viability. Density (streams per rack unit or per VM), watts-per-stream, and supportability affect both capital and operational expenditure for 24/7 news services. A platform that requires twice the rack space or power budget for the same stream count imposes compounding costs over a multi-year deployment.
Technical specifications and trade-offs
When evaluating SRT platforms for newsroom ingest, the following dimensions should appear in any structured comparison:
-
Form factor and density: Chassis rack units, module slots, blade capacity—critical when consolidating many decoders and encoders in a master control room.
-
Aggregate SRT throughput: The total Gb/s of SRT traffic the node can sustain under encryption.
-
Concurrent connection count: Maximum simultaneous SRT sessions at specified bitrates.
-
Latency tuning range: Minimum configurable latency (ms) and recommended operating buffer for given loss profiles.
-
Redundancy modes: Dual-path SRT, ST 2022-7, hot/hot chassis, hot/standby appliance failover.
-
Integration APIs: NMOS IS-04/IS-05, MOS, REST, SNMP, or cloud-native IAM and orchestration APIs.
-
Codec and timing support: AVC, HEVC, JPEG XS encode/decode; PTP/PTPv2 for frame-accurate timing.
The table below compares representative platforms across these dimensions. Figures are drawn from published product specifications; validate with current vendor documentation before procurement decisions.
|
Platform / approach |
Typical form factor |
Representative density / throughput |
Concurrent connections (example) |
Redundancy and failover options |
Typical newsroom use case |
|---|---|---|---|---|---|
|
Appear X20 (hardware-accelerated SRT) |
2 RU modular chassis |
Up to 72 Gb/s SRT per chassis; high streams-per-RU via modular cards |
Up to ~1,536 per 2 RU chassis |
Hardware-offloaded AES and SRT; PTP, ST 2022-7, module redundancy, centralised estate management (XM) |
Centralised high-density hub ingest and interfacility links |
|
Haivision Makito X4 (appliance / blade) |
Blade / rack chassis (1 RU–4 RU scale) |
Up to 84 HD or 21 UHD encoder channels in a 4 RU, 21-blade chassis |
Counts encoder channels, not SRT connections; per-blade limits apply |
SRT-native; Haivision SRT Gateway and Hub cloud options available |
Edge-to-core encoding, distributed encoder estates, SRT gateway use |
|
Net Insight Nimbra (media network node) |
Modular rack nodes (e.g., 3 RU) |
Nimbra 1060: up to 1.2 Tb/s of guaranteed capacity in 3 RU |
Designed for large-scale media WAN; connection scaling by node type |
Carrier-grade QoS, multi-path transport, service-provider-grade redundancy |
Interfacility WAN backbones and high-scale primary distribution |
|
AWS Elemental MediaConnect (cloud service) |
Managed cloud service (per AWS region) |
Up to 2.5 Gbps aggregate on a large flow; 400 Mb/s aggregate output is the standard design point |
Horizontal scaling; default quotas of 20 flows per Region and 50 outputs per flow |
Cloud-region redundancy, multi-flow outputs; relies on availability zones |
Elastic, short-term ingest; cloud-native processing and distribution |
These rows measure different things. Appear’s figures count SRT gateway connections and throughput. Haivision’s count encoder channels, Net Insight’s is capacity on a guaranteed-bandwidth WAN platform, and AWS’s is aggregate flow throughput. Use the table to compare form factor and scaling model, and compare like with like only after asking each vendor for connection counts and throughput at your bitrates and latency settings.
Trade-off guidance
A few rules of thumb help narrow the field:
-
When a single ingest node must handle more than 1–2 Gb/s of encrypted SRT traffic or dozens of concurrent full-frame UHD feeds, hardware-accelerated platforms offer predictable pps performance and lower watts-per-stream.
-
When usage is highly elastic or event-driven—a one-off election night or a temporary bureau—cloud and software approaches provide cost flexibility, but require orchestration to prevent per-instance CPU and pps exhaustion.
-
For sub-second latency targets, specify minimum buffer and latency settings explicitly and test under expected packet-loss and jitter conditions. Higher packet loss forces larger buffers or stronger FEC, which directly increases end-to-end delay.
Related approaches and standards
SRT does not exist in isolation. Several adjacent protocols and standards interact with or complement SRT-based ingest architectures:
-
RIST and Zixi are alternative reliable-transport protocols frequently benchmarked against SRT for error-recovery behaviour, latency characteristics, and installed-base breadth.
-
ST 2110 is the family of SMPTE standards for uncompressed IP-based production (timing, video essence, audio, ancillary data). SRT gateways commonly bridge compressed contribution feeds into ST 2110 production domains.
-
ST 2022-7 defines seamless protection switching for dual-path IP streams and is often used alongside SRT for hitless redundancy across diverse network paths.
-
NMOS IS-04 and IS-05 provide discovery and connection management for IP-native production facilities, enabling SRT gateways to register and be controlled within larger automated systems.
-
JPEG XS is a visually lossless, ultra-low-latency codec increasingly paired with SRT for contribution links where bandwidth is constrained but latency and quality demands are high.
-
PTP (IEEE 1588) supplies frame-accurate timing across production sites, essential when SRT-delivered feeds are integrated into multi-camera or live-mixing timelines.
How Appear addresses this
The X20 runs SRT in dedicated hardware. Each accelerated SRT module in an X10 or X20 carries up to 128 concurrent connections, and an X20 in 2RU takes 12 modules: 1,536 connections and up to 72 Gbit/s of SRT traffic in one chassis, at the central point where bureau and ENG feeds converge. X5 is a compact half-rack unit with accelerated SRT for smaller sites, at up to 16 inputs and 16 outputs. Modules support caller, listener, and rendezvous modes, with AES-128 or AES-256 encryption.
The 72 Gbit/s figure is a ceiling, and this article’s sizing advice applies to it. The retransmit buffer is shared across a module’s outputs, so total capacity falls as the configured maximum latency rises: up to 6 Gbit/s per module at a 1 s maximum, and less as the window widens. Sub-second windows, the normal case for breaking-news links, sit inside that figure. Latency is set on each SRT input between 60 ms and 8 s, so a live two-way and a long-haul bureau link can run different settings on the same chassis.
For redundancy, ST 2022-7 seamless protection merges on the SRT module in the same buffer as ARQ, so protection adds no latency. SRT broadcast bonding duplicates a stream over two connections and merges inside the receive buffer, and RTP-cloned redundancy is also available. Each protected stream uses two of the module’s 128 connections. L2 link bonding with the SRT image runs in active-backup mode on IPx100 and IPx110, which gives path failover rather than aggregated capacity; X5 does not support it.
FEC sits on the UDP/RTP side. Appear’s SRT recovers loss by ARQ and carries no FEC filter, which matches this article’s guidance to run small ARQ windows with minimal or no FEC on breaking-news links. For long-haul bureau links, where the guidance above calls for stronger FEC, the equivalent levers are a wider latency window and a second path. ST 2022-1 FEC is a licensed option on the IP gateway path for UDP/RTP feeds, and on X5 it applies to output only.
Configuration and status are available through a web interface and a JSON-RPC API, so an automation layer can create, change, and remove SRT inputs and outputs. SNMP v2/v3 alarms, syslog alarm forwarding, and a Prometheus endpoint cover monitoring. SRT inputs and outputs publish their own metrics there, including retransmit and drop counters, and RTT appears in the interface. A sender-side drop counter shows that a receiver asked for packets too late, but only receive-side metrics confirm loss on the link, so monitor both ends. Priority pre-emption and MOS integration are handled by the newsroom’s automation layer, driving the chassis through the API. XM adds inventory, alarms by severity, software and licence lifecycle, and configuration export across deployed chassis.
Where ingest feeds an ST 2110 facility, the switch modules take PTP (SMPTE ST 2059-2 and ITU-T G.8275.2 profiles) or genlock as the clock reference, and the ECx210 module decodes AVC and HEVC to ST 2110. Codecs are AVC and HEVC on the ECx modules and JPEG XS on the SIx family. The SRT modules have 1/10G ports, and 25, 40, and 100G come with the ST 2110 and 100G switch modules.
VX covers elastic or event-driven ingest. It runs SRT in software on commodity Linux, private-cloud VMs, or public-cloud instances, with SRT RTT, retransmission, and lost-packet metrics in Prometheus and Grafana. It recovers loss by ARQ without FEC.
As with any vendor, ask for a load test at your feed count, bitrates, latency settings, and encryption before committing.
Appear delivers the low-latency contribution, processing and transport behind the world's most demanding live productions.
