Performance and reliability analysis of distributed computing architectures under heavy loads

A Project of the D'Agents Laboratory at Dartmouth College

Daniel Bilar

Standards in distributive computing architecture, such as OMG's Corba, Sun's Jini or Microsoft's DCOM, are slowly gaining ground. The architectures perform well under small loads (i.e a few dozen objects and invocations,), but their characteristics under real-life heavy loads (1000's of objects and invocations) have yet to be investigated.

Three issues, among others, ought to be investigated:

1) Performance: Does the system effectively scale to thousands of objects in terms of lookup time of objects, server respose time and load balancing?

2) Fault reliability: Do neuralgic points exist which jeopardize the working of the system under heavy loads? How does performance degrade when inevitable disturbances are introduced?

3) Fault recovery: Are there effective mechanisms in place to ensure automatic fault recovery in the case of system failures or massive degradation?


Preliminary results

Corba experimental results

Relevant Dartmouth papers:

Mobile Agents for Distributed Information Retrieval: Brian Brewington, Robert Gray, Katsuhiro Moizumi, David Kotz, George Cybenko and Daniela Rus. In Matthias Klusch, editor, Intelligent Information Agents, chapter 15, Springer-Verlag, 1999.

Network Awareness and Mobile Agent Systems: Wilmer Caripe, George Cybenko, Katsuhiro Moizumi and Robert Gray. IEEE Communications Magazine, 36(7):44-49, July, 1998.

Matching Conflicts: Functional Validation of Agents:George Cybenko and Guofei Jiang. In AAAI Workshop of Agent Conflicts, pages 14-19, Orlando, Florida, July, 1999. AAAI Press.

Relevant projects, inside D'Agents:

Cybenko: Networking sensing, navigation and resource discovery tools

Relevant projects, outside D'Agents:

Performance of CORBA Objects

Research on high performance CORBA