CAREER: Algorithms and Tools for Fault Tolerance and Migration in Distributed Computing Environments

这个研究项目集中在容忍资源动态性的两个重要领域：体系结构独立检查点（AIC）和用于高可用性的原语。与体系结构无关的检查点是一种以与机器无关的方式保存程序执行状态的能力，以便可以在不同的处理平台上进行恢复。这可能是在不同的机器上，或者对于并行程序，不同的并行处理环境。PI从不同的角度攻击AIC。这些是ScaLAPACK中特定于应用程序的AIC， C和Fortran程序的半透明到透明的AIC， Java应用程序的AIC，以及高性能Fortran应用程序的AIC。结果将是一组用于各种广泛使用的计算工具的AIC的方法和实现。该研究项目的第二部分是设计和实现三种当前流行的分布式计算工具（ScaLAPACK、MPI和Java）中的高可用性原语。目前，每种方法的支持都是有限的。不允许出现故障、处理器撤销和后续可用性、网络分区和计划关闭等情况。PI计划解决这些缺陷。其结果将是为分布式计算提供更健壮的工具，以及可能被纳入未来工具设计的方法。

This research project focuses on two important areas in tolerating resource dynamicism: architecture- independent checkpointing (AIC) and primitives for high availability. Architecture-independent checkpointing is the ability to save the execution state of a program in a machine independent manner so that it may be restored on a different processing platform. This could be on a different machine, or for a parallel program, a different parallel processing environment. The PI is attacking AIC from different angles. These are applicationspecific AIC in ScaLAPACK, semi-transparent to transparent AIC for C and Fortran programs, AIC of Java applications, and AIC of HighPerformance Fortran applications. The result will be a collection of methods for and implementations of AIC for a variety of widely used computing tools. The second portion of this research project is to design and implement primitives for high availability in three currently popular tools for distributed computing: ScaLAPACK, MPI, and Java. Currently, each is limited in its support. Occurrences such as failures, processor revocation and subsequent availability, network partitions, and planned shutdowns are not accommodated. The PI plans to address these deficiencies. The result will be more robust tools for distributed computing, plus methodologies that may be incorporated into the design of future tools.