Technical Design And Implementation Of Process Migration Based On Distributed Linux System

Distributed system has become one of the best methods to utilize existing idle computing resources to handle difficult tasks. But in order to make full use of the resources in distributed systems, with good load-balancing, high communication performance and high availability, process migration is required.Distributed Operating Systems involve many aspects in Computer Science, e.g.: Process Management and Scheduling, Inter-process Communication, Storage Management, Parallel Control, File Systems, Transaction, Clock Synchronization, Security Mechanism, etc. The research topic of this thesis is "Load Distribution", a destination of distributed process management and scheduling.The goal of "Load Distribution" is high efficient usage of resources in distributed systems. This goal can be described as "Load Balancing", which means trying best to average the workload to the whole system; or "Load Sharing", which means only trying to help those overloaded resources. The former is much more difficult to attain than the latter. We discuss "Load Balancing" in our thesis.The implementation method of load balancing is "Process Migration", that is, replacing a process and all its relative status information on another processor. In fact, in the way of moving processes from overloaded system to a lower-loaded system, workload can be balanced, and the whole performance of the distributed system is approved.In order to implement load balancing, we need do work of three aspects: gathering and monitoring load information of each processor in distributed systems in order to decide whether or not move some load on some processor; monitering the information of processes in order to choose suitable candidate process to migrate and where to, at suitable time; migrating process.This paper discusses the design and implementation of three modules for the process migration in distributed systems based on Linux, which are "Load Information Management", "Distributed Scheduling" and "Process Migration Mechanism".We give a new algorithm for dynamic process migration after comparing the existing algorithms; offer the system users a transparent distributed-OS interface by modifying Linux kernel and adding a Global Operation module; design a Migrate Protocol so that the Client and Server can give the consistent view of process and transparent remote execution, and handle link failure and node failure. The success of the implementation is proved by a test using parallel compilation task and the system's performance is compared with another two similar systems.