Development of a responsive operating system for scalable multiprocessor systems

点击翻译按钮获取中文摘要

We designed and implemented an operating system which supports parallel real-time processing in scalable multiprocessor systems.l. We proposed a scalable load balancing algorithm based on overlapped domain decomposition and compare its performance with conventional algorithms by using simulation. We showed the proposed algorithm is very effective and scalable when choosing a domain size appropriate to the interconnection network topology and the communication cost. Various aspects of load balancing policies such as information exchange, migration decision, node selection, etc.are also discussed.2. We proposed global page placement policies in distributed memory parallel systems. Global page placement policies provides an alternative to the expensive disk swapping present in conventional virtual memory systems by taking advantage of faster internode memory access. In support of global page placement, selection of a page for replacement, selection of a node to which a page should be sent, and so on.3. We designed and implemented an operating system that supports light-weight objects. We also designed the distributed processing facilities such as remote object invocation, global naming, and object replication/migration, etc., which are necessary for responsive processing.