Studies on Dependable Distributed Storage Cluster for SoC Design System using SAN/NAS

本课题研究的主要目的是研究和开发可靠的计算技术，为基于SAN/NAS的SoC设计系统构建高可靠的分布式存储集群。我们重点研究了一些元器件技术，取得了以下研究成果。考虑到SAN/NAS正在广泛分布的情况下，我们研究了分布式系统中的检查点和恢复方案。首次提出了基于增量快照的非协调检查点和协调检查点的混合方案。对非协调检查点和日志记录的混合方案进行了二次研究。新思想的引入使我们能够推广前人研究的评价模式。针对高可靠通信系统，研究了基于卷积码的可靠文件传输技术。提出了一种基于(n，k，m)卷积码的分组级前向纠错方案，并对其分组可恢复性进行了评估，计算了编码或译码量。此外，还对FEC进行了如下工作：分组缓存大小受限时的可恢复性分析；混合ARQ方案的重传协议；生成矩阵密度与重构后接收速率的关系等。在IP版本维护采用分布式存储集群的假设下，讨论了读写操作的访问协议。混合数据复制协议被概括为我们所考虑的梯形协议。在协议中进一步引入了概率仲裁系统。最后对SoC多代IP版本的多代保护协议进行了探讨。

The main purpose of our study has been the research and development of dependable computing techniques in order to construct the highly reliable distributed storage cluster for SoC design system using SAN/NAS. We focused on some element technology to obtain the following study results.1. Considering the situation that SAN/NAS is spreading over wide areas, we investigated the checkpointing and recovery schemes in distributed systems. We first presented the hybrid scheme between uncoordinated checkpointing and coordinated one by using the incremental snapshot. We secondary study the hybrid scheme between uncoordinated checkpointing and logging. Introducing the new ideas enabled us to generalize the evaluation model of a prior study.2. For highly reliable communication systems, convolutional code based dependable file transfer technique was examined. We presented (n, k, m) convolutional code based packet level FEC scheme and evaluated its packet recoverability and calculated amounts of encoding or decoding. Moreover, some works for the FEC were carried out : the recoverability analysis where the packet buffer size was restricted to finite ; retransmission protocols for hybrid ARQ scheme ; the relation between the generator matrix density and the post-reconstruction receiving rate ; and so on.3. On the assumption of distributed storage cluster for IP version maintenance, access protocols for read/write operations were discussed. The hybrid data replication protocol was generalized down to the trapezoid protocol we considered. The probabilistic quorum system was further introduced to our protocol. We finally inquired into the multi generational conservation protocol for SoC multi-generational IP version.