摘要： 为实现协同计算任务的透明化设计部署及在并行计算中解耦合，设计了多虚拟机协同计算任务的分发部署及运行框架. 该框架利用隐式通信简化了进程间的交互，为用户提供一种高可靠性和容错的计算环境. 在框架设计方案中基于虚拟器件及应用程序虚拟化支撑技术，提出了六方面的任务分发部署和运行关键技术，包括可执行文件包描述及其提交方法、可执行文件的自动化分发部署方法、任务启动与加载方式、特征单元状态变迁、采用的消息通信原语、系统容错性方法等. 通过这些关键技术实现了所提出的任务的分发部署及运行框架，并进行了性能测试分析. 实验结果表明，设计的框架和采用的关键技术提高了任务的分发部署速度及系统吞吐量，提高了资源利用率，提高了任务的加速比和系统的运行效率.

A framework based on a three-layer structure for collaborative computing task distribution, deployment and execution over multiple virtual machines has been designed to implement transparent design and placement of tasks. The framework simplifies the interaction among processes with implicit communication,and provides a high reliable and fault-tolerant computing environment. Based on the supporting technologies including virtual appliance technology and application virtualization, six key techniques are proposed for the framework design. They are description of executable files packages and their submission methods, automatic distribution and deployment of executable files, the way of starting and loading tasks, status transition of feature units, message communication language, and system fault tolerance approaches. Based on these key techniques, a prototype system has been developed, and its performance tested. The experimental results show that the framework and key techniques can improve the speed of tasks distribution, deployment and execution, and the system throughput. The task speedup rate is therefore increased.