公开(公告)号：US06990668B1

公开(公告)日：2006-01-24

申请号：US09421585

申请日：1999-10-20

申请人： Robert Miller ,  Vicki Lynn Morey ,  Kiswanto Thayib ,  Laurie Ann Williams

发明人： Robert Miller ,  Vicki Lynn Morey ,  Kiswanto Thayib ,  Laurie Ann Williams

IPC分类号： G06F9/461

CPC分类号： H04L29/06 ,  H04L1/1628 ,  H04L69/12

摘要： An apparatus and method passively determine when a job in a clustered computing environment is dead. Each node in the cluster has a cluster engine for communicating between each job on the node and jobs on other nodes. A protocol is defined that includes one or more acknowledge (ACK) rounds, and that only performs local processing between ACK rounds. The protocol is executed by jobs that are members of a defined group. Each job in the group has one or more work threads that execute the protocol. In addition, each job has a main thread that communicates between the job and jobs on other nodes (through the cluster engine), routes appropriate messages from the cluster engine to a work thread, and signals to the cluster engine when a fault occurs when the work thread executes the protocol. By assuring that a dead job is reported to other members of the group, liveness information for group members can be monitored without the overhead associated with active liveness checking.

摘要翻译： 设备和方法被动地确定群集计算环境中的作业何时死亡。 集群中的每个节点都有一个集群引擎，用于在节点上的每个作业和其他节点上的作业之间进行通信。 定义了包括一个或多个确认（ACK）轮，并且仅在ACK轮之间执行本地处理的协议。 协议由作为定义组成员的作业执行。 组中的每个作业都有一个或多个执行协议的工作线程。 此外，每个作业都有一个主线程，通过其他节点（通过群集引擎）在作业和作业之间进行通信，将适当的消息从群集引擎路由到工作线程，并在发生故障时向群集引擎发出信号 工作线程执行协议。 通过确保向组织的其他成员报告死亡的工作，可以监视组成员的活动信息，而不需要与主动活动检查相关的开销。