新型互联网应用的蓬勃发展对端到端网络传输服务的性能提出了更高的需求，如更低的端到端时延、更高的传输吞吐量。传输过程中的丢包恢复会导致传输性能的下降。面对丢包问题，越来越多的传输服务提供商使用前向纠错编码来恢复高性能传输任务中丢失的数据包。基于前向纠错编码的丢包恢复是一种以冗余带宽资源和额外计算开销换取传输性能提升的技术。然而该技术在面对高性能传输需求时仍存在诸多挑战需要克服。具体来说，不准确的编码率决策、计算引入的编解码处理时延以及不合理的机会链路传输策略会降低传输的性能。本文致力于解决上述问题以优化基于FEC丢包恢复机制的高性能传输的时延及吞吐性能。本文的主要研究内容及贡献如下：1. 针对编码率难以准确决策导致尽力而为的FEC解码成功率无法达到预期目标的问题，本文对现有解码成功率保障方案进行了测量，并根据测量得到的观察提出了面向低时延承诺的解码成功率保障方案。该方案结合了段内及段间编码，通过保守地制定段内编码率及合理地补充段间编码包实现了解码成功率和带宽开销之间的良好权衡。实验显示，该方案可以在带宽浪费仅为理论保障方案的1%-16%的情况下实现目标解码成功率。此外，该方案可与轻量级丢包预测算法良好地配合，这使得它在计算资源有限的环境中依然可以发挥作用。2. 针对现有编解码方案在多跳传输中累积的处理时延过大的问题，本文综合考虑了处理时延和系数带宽开销，提出了面向多跳低时延传输的编解码加速技术。该技术采用预先协商的映射表及其下标传输系数矩阵信息，以实现较低的系数开销。同时，它通过在中继节点上流式地对数据包进行重新编码来减少处理延迟。实验显示，与传统的里索码及随机线性重编码相比，本文所提方案可以将处理延迟降低高达88%，而不增加系数带宽开销。3. 针对机会链路难以准确预测继而导致编码包传输吞吐量下降的问题，本文对机会链路预测的时间特性进行测量并提出了不完美机会链路预测环境下高吞吐传输调度方案。该方案跟踪所使用的预测算法的准确性，自适应地确定每次传输调度的时间窗口大小，以尽量减少调度引入的不准确路径。同时，该方案使用了针对移动目标传输特征的启发式原则，用以解决多个请求之间的资源竞争问题。实验表明，与现有传输调度方案相比，本文所提方案的吞吐量提高了约一倍。
The vigorous development of new Internet applications has put forward high requirements for the performance of end-to-end network transmission services, such as low end-to-end latency and high transmission throughput. Packet loss recovery during transmission will lead to a decrease in transmission performance. Facing the problem of packet loss, more and more transmission service providers use Forward Error Correction (FEC) technology to recover lost packets in high-performance transmission tasks. FEC-based packet loss recovery is a technique that trades redundant bandwidth resources and computational overhead for improved transmission performance. However, there are still several challenges that need to be overcome when FEC technique is used for high-performance transmission. Specifically, inaccurate decisions of the FEC code rate, the introduced FEC processing delay, and unreasonable transmission scheduling decisions on opportunistic links will degrade the performance of the transmission. This thesis aims to address the aforementioned issues to optimize the latency and throughput of high-performance transmission based on FEC packet loss recovery mechanisms. The main research contents and contributions of this thesis are as follows:1. To address the problem that the decoding success rate is difficult to achieve the expected goal due to the inaccurate code rate determination, this thesis measures existing solutions that may achieve the expected decoding success rate, and based on the measurement results, this thesis proposes a decoding success rate guarantee scheme for low latency commitment. This scheme combines intra- and inter-generation coding, and achieves a good trade-off between decoding success rate and bandwidth overhead by conservatively determining the code rate of intra-generation coding and appropriately supplementing redundant packets of inter-generation coding. Experiments show that this scheme can achieve the targeted decoding success rate while its bandwidth waste is only 1%-16% of the FEC with the code rate that can theoretically guarantee the target success rate. Furthermore, this scheme can work well with computationally lightweight prediction algorithms, which makes it extremely useful for the transmission environment with limited computing resources.2. To address the problem of excessive processing delay accumulated in multi-hop transmission, this thesis comprehensively considers the processing delay and coefficient bandwidth overhead, and proposes a cost-efficient FEC scheme for low-latency multi-hop transmissions in overlay networks. This scheme achieves a low coefficient overhead by using a pre-negotiated mapping table and its seed to transmit the information about the coefficient matrix. At the same time, it reduces FEC processing delay by recoding packets in an on-the-fly manner at relay nodes.Real-world experiments demonstrate that this scheme can reduce the processing delay by up to 88% without increasing the coefficient overhead compared to Reed-Solomon code and RLNC recoding schemes.3. To address the problem of transmission throughput degradation caused by inaccurate opportunistic link prediction, this thesis measures the time characteristics of opportunistic link prediction and proposes a high-throughput transmission scheduling scheme in imperfect opportunistic link prediction environments. The scheme tracks the accuracy of the prediction algorithm used and adaptively determines the time window size for each transmission schedule to minimize the use of inaccurate opportunistic links. At the same time, this solution uses a heuristic principle based on the transmission characteristics of moving targets to solve the resource competition problem between multiple requests. Evaluation results show that the throughput of the proposed method is about doubled compared to the existing scheduling algorithm.
