PRRT: Predictably Reliable Real-time Transport

Reliable transport layer Internet protocols do not satisfy the requirements of packetized, real-time multimedia streams. First, major limitations result from their primary design objective of serving total reliability without tolerating residual packet loss. This property leads to unpredictable delivery delay on lossy network paths and conflicts with the strict rendering deadlines of multimedia services that explicitly prefer timeliness over reliability. Second, the strict layering of the ISO/OSI network stack prevents applications to communicate their specific quality of service (QoS) requirements to the transport layer. Consequently, transport protocols do not provide an interface for the negotiation of constraints on packet loss and delivery delay. Third, as the provision of scalable one-to-many transport requires careful design – especially under combination with error control – it is insufficiently supported by reliable protocols. Yet broadcast or multicast distribution of digital media is efficient and not unusual. As of today these issues are clearly unsolved in the prevalently HTTP/TCP-based media streaming such that the available Internet bandwidth is significantly underutilized and the presentation quality suffers severely.

We define predictable reliability (see Figure 1) as a novel, capacity-approaching transport paradigm, supporting an application-specific level of reliability under a strict delay constraint. This paradigm is being implemented into a new protocol design – the Predictably Reliable Real-time Transport protocol (PRRT). The protocol combines the fundamental concepts of proactive and reactive packet-level error control into an adaptive hybrid error coding architecture. The flexibility of the hybrid scheme enables the protocol to adaptively follow the dynamic capacity of the packet-erasure channels generated by a wide range of Internet protocol infrastructures. Combined with packet loss notifications via negative acknowledgments, it provides capacity-approaching coding efficiency in point-to-point as well as one-to-many transmission scenarios.

Figure 1: Predictable Reliability

In order to predictably achieve the desired level of reliability, proactive and reactive error control must be optimized under the application's delay constraint. Hence, predictably reliable error control relies on stochastic modeling of the protocol's reaction to the network path's packet loss behavior. A block-erasure model captures the characteristics of the packet loss process. Further, a protocol performance model is being developed that predicts the protocol's residual packet loss rate as well as its coding overhead based on the statistical representation of the network state. The performance model reflects the efficiency of one-to-many error control and incorporates the impact of unreliable delivery of the negative acknowledgments. The result of the joined modeling is periodically evaluated by a reliability control policy that validates the protocol configuration under the application constraints and under consideration of the available network bandwidth. The adaptation of the protocol parameters is formulated into a combinatorial optimization problem that is solved by a fast search algorithm incorporating explicit knowledge about the search space.

Experimental evaluation of PRRT in real Internet scenarios demonstrates that predictably reliable transport meets the strict QoS constraints of high-quality, audio-visual streaming applications. In particular, broadcast services over Internet Protocol require packet streams to be delivered at a residual loss rate of 10-6 to 10-5 under a delay constraint of few hundred milliseconds, depending on their degree of interactivity. Within different experiments, the protocol implementation has been evaluated at the transport of high-quality broadcast TV via Internet Protocol. Especially wired wide area network paths as well as wireless and mobile networks expose the transport protocol to highly dynamic packet loss rates and propagation delays. Comparative experiments with recent advancements in dynamic Internet video streaming confirm PRRT's significant gain in efficiency. The fairness towards existing transport protocols on shared network paths is demonstrated under delay-based congestion control.

Theoretical Foundation

The core of our media-oriented transport protocol is an Adaptive Hybrid Error Correction (AHEC) approach based on previous work from Guoping Tan (Application Layer Hybrid Error Correction Techniques for DVB Services in Wireless Home Networks). Adaptive coding is the consequent usage of the bidirectional characteristic of available IP networks. The highly flexible composition of NACK based ARQ and adaptive packet-level FEC leads to near-optimal coding efficiency as it is controlled by analytical parameter derivation based on a statistical channel model. 

Related Publications

Contact Person