Pular para o conteúdo principal

[EN] SRT: The Engineering of Low Latency and Packet Recovery


The SRT (Secure Reliable Transport) protocol transformed the public internet into a viable medium for professional video contribution by replacing the paradigm of unpredictable transport with deterministic latency and intelligent recovery. While TCP-based protocols suffer from Head-of-Line Blocking and pure UDP ignores data integrity, SRT utilizes a selective retransmission mechanism that guarantees the delivery of every frame without interrupting the continuous streaming flow, even under adverse network conditions.

Knowledge Architecture: Study First

To master the fine-tuning and engineering of SRT, it is imperative to understand the pillars that support its technology stack:

Concept: NAK Recovery and Fixed Latency

Unlike TCP, which confirms every received packet (ACK), SRT operates primarily via NAK (Negative Acknowledgement). The receiver monitors packet number sequences in real-time; upon detecting a gap, it immediately sends a "packet not received" signal. The secret of its resilience lies in the configurable latency buffer: the receiver holds packets for a precisely defined time, creating a window of opportunity for retransmissions requested via NAK to arrive and be reordered before playout.

Operation and Internal Structure: SRT

SRT's internal mechanics overlay UDP with a layer of congestion control and high-precision timestamping.
  • TSBPD (Time-Stamped-Based Packet Delivery): Each packet carries a timestamp from the encoder. The receiver uses this data to deliver packets to the application at exactly the same rhythm they were generated, eliminating network jitter.
  • Selective Retransmission: The transmitter keeps a copy of recently sent packets in memory (send buffer). When it receives a NAK, it resends only the missing packets, avoiding the overhead of retransmitting data that has already reached the destination.

Calculating and Adjusting Minimum Latency

Adjusting latency is a video engineer's most critical decision. The value must be calculated based on the network's RTT. The gold standard formula is: Latency = RTT * Multiplier. For stable professional networks, a multiplier of 3 to 4 is used. On unstable or long-distance networks (packet loss above 5%), 5 to 8 is recommended. Why this calculation? A complete retransmission cycle (Detect loss -> Send NAK -> Transmitter receives NAK -> Retransmitted packet arrives) consumes exactly 1 RTT. Setting latency to 4x RTT ensures the protocol has at least four attempts to recover the same packet before playout time expires. If your RTT is 100ms, your safe latency should be at least 400ms. Adjusting below this on jittery networks will result in dropped frames and image artifacts.

To learn more about the subject:


Technical Disclaimer and Intellectual Property Notice

This blog presents analyses and facts based exclusively on technical documentation, RFCs, and publicly available materials on the global computer network. The information contained herein is compiled for strictly educational and technical reference purposes.

Lack of Affiliation: This project is independent and has no official affiliation, endorsement, or link with the developers, companies, or rights holders of the mentioned technologies. All trademarks and logos cited belong to their respective owners.

Liability: The implementation of any protocol or configuration based on these notes is the sole responsibility of the user. The author disclaims any liability arising from the misuse of this information.

Rights and Corrections: We fully respect intellectual property. If you are the rights holder of any material or technology cited here and identify the need for corrections, adjustments, or wish to make official comments, please send a private message directly to the author for immediate resolution.

Comentários

Postagens mais visitadas deste blog

[PT] TCP: O Arquiteto da Confiabilidade em Redes de Dados

Enquanto o Protocolo de Internet (IP) é frequentemente comparado ao sistema de endereçamento de envelopes, o Transmission Control Protocol (TCP) é o serviço de correio registrado que garante que o conteúdo não apenas chegue ao destino, mas chegue na ordem correta e sem corrupção de dados. Em uma rede inerentemente não confiável e baseada em melhor esforço, o TCP atua como a camada lógica que transforma o caos da comutação de pacotes em um fluxo contínuo e ordenado de informações. Ele é um protocolo orientado à conexão, o que significa que antes de qualquer dado ser transmitido, uma sessão formal deve ser estabelecida e mantida entre as duas extremidades. Pré-requisitos e Contexto Técnico Para compreender profundamente o funcionamento do TCP, é recomendável que o leitor esteja familiarizado com os conceitos de endereçamento e roteamento do IP (Internet Protocol) , conforme explorado em nossas publicações anteriores. O TCP opera sobre a camada IP, adicionando a inteligência de contro...

[ EN ] OSPF: The Mathematical Rigor of Link-State Routing Efficiency

[ EN ] OSPF: The Mathematical Rigor of Link-State Routing Efficiency OSPF stands as the deterministic heart of modern enterprise networks, utilizing the Dijkstra algorithm to transform raw link data into a loop-free topology of shortest paths. While distance-vector protocols rely on second-hand information, OSPF (Open Shortest Path First) demands a complete, synchronized map of the entire area, ensuring that every routing decision is based on an absolute global truth rather than neighbor-based rumors. Knowledge Architecture Study First Genesis and Historical Context Internal Functioning and Structure OSPF At the core of OSPF lies the Shortest Path First (SPF) algorithm, also known as Dijkstra's algorithm. To understand OSPF, one must understand that it does not simply "exchange routes"; it exchanges Link-State Advertisements (LSAs). These LSAs describe the state of every interface, the cost associated with it, and the neighbors connected to it. These advertisements are...

[ PT ] OSPF: A Engenharia de Estado de Enlace e a Eficiência do Algoritmo de Dijkstra

[ PT ] OSPF: A Engenharia de Estado de Enlace e a Eficiência do Algoritmo de Dijkstra O Open Shortest Path First (OSPF) é a espinha dorsal da conectividade dinâmica em redes corporativas, utilizando a inteligência do estado de enlace para garantir que cada roteador possua um mapa completo e sincronizado da topologia. Ao contrário de protocolos baseados em vetores de distância, o OSPF não confia cegamente no que seus vizinhos dizem, mas sim no que eles veem, processando essas informações através do rigor matemático do algoritmo de Dijkstra para determinar o caminho mais curto e eficiente para o tráfego de dados. Arquitetura de Conhecimento Estude Antes Funcionamento e Estrutura Interna OSPF Hello 10s / Dead: 40s (em redes Broadcast) Para aprender mais sobre o assunto [Clique aqui para investigar] a documentação oficial da RFC 2328 para OSPFv2. [Clique aqui para investigar] as diferenças detalhadas entre todos os tipos de LSAs e áreas Stub. [Clique aqui para investigar] como o OSPF...