Controlled Recovery in Distributed Systems

提案编号CCR-9901391分布式系统中的受控恢复PI：Sukumar Ghosh分布式系统传统上由一组互连的进程组成，这些进程不共享全局内存。进程之间的通信要么通过本地共享的内存，要么通过沿着连接进程对的链路传播的消息。在许多应用中，分布式系统需要从不可预见的性质的故障或从环境扰动中自发地运行时恢复。这类系统的设计，称为自稳定系统或简称稳定系统，近年来得到了广泛的研究。当前稳定分布式系统的一个主要弱点是它们无法将恢复时间与故障程度关联起来。第二个令人担忧的问题是，在恢复阶段，没有故障的进程可能会受到污染，这一问题可能会对大型分布式系统的迅速扩散产生令人担忧的后果。这激发了分布式系统中受控恢复的研究，在分布式系统中，控制意味着恢复过程中满足一定的性能约束。对受控恢复的研究将有助于更好地理解安全关键系统的设计。除了研究这种系统设计的算法基础外，还将研究在多媒体数据路由和移动节点自组织网络领域的应用。

Proposal Number CCR-9901391Controlled Recovery in Distributed Systems PI: Sukumar GhoshA distributed system traditionally consists of a set of interconnected processes that do not share a global memory. Communication among processes is either through locally sharedmemory, or through messages that propagate along the links joining pairsof processes. In many applications, a distributed system is required to stage a spontaneous run-time recovery from failures of an unforeseen nature, or from environmental perturbations. The design of such systems, known as self-stabilizing systems or simply stabilizing systems, has beenextensively studied in recent years. A major weakness in current stabilizing distributed systems is their inability to correlate the recovery time with the extent of failure. A second concern is the potential of non-faulty processes being contaminated during the recovery phase, a problem that may have alarming consequences with the rapid proliferation of large distributed systems. This motivates the study of controlled recovery in distributed systems, where controlimplies that certain performance constraints are satisfied during recovery.The study of controlled recovery will lead to a better understandingof the design of safety-critical systems. In addition to studyingthe algorithmic foundations of the design of such systems, applications in the areas of multimedia data routing and ad-hoc networks of mobilenodes will be examined.