Initial implementations of middleware based on standards such asCORBA have concentrated on host and language transparency issues inorder to demonstrate interoperability. They have largely adopted ano-replication approach and have frequently neglectedperformance-at-scale issues. This has led to continuing deployment ofeither non-scalable full-replication approaches or ad-hocmessaging-based middleware for applications such as intelligentnetworks, WWW applications and collaborative virtual reality. Theseapplications require millions of objects globally distributed acrosshundreds of hosts and demand a very high throughput of low-latencymethod invocations. Our main research aim is to be able to reason aboutthe performance of such applications when using scalablepartial-replication and object-oriented approaches to middleware. Ourapproach is to use a simulator to explore potential design andimplemention choices. Our current simulator-driven design, called“MinORB”, has been fully implemented and tested. MinORBsupports scalable high performance by a combination of techniques,including weak and application-specified consistency and partialreplication using fine-grained proxy caching. Experimental results showthat our work compares very favourably with other leadingimplementations, such as OmniORB. Scalability is unparalleled, with upto 1,000,000,000 objects per address space, a maximum throughput of42,000 invocations per second and service times as low as 4 ms
展开▼
