Login
Computer Science @ Rochester
Thursday, May 02, 2002
11:00 AM
CSB 209
Ph.D. Thesis Proposal
Chunqiang Tang
University of Rochester
Distributed Computing in Heterogeneous Environments
With the rapid proliferation of the Internet, more and more applications are developed or ported to it in order to take advantage of resources only available at distributed sites or to address to a broader audience. One inherent property of the Internet is its heterogeneity in terms of architecture, operating system, programming language, computation power, network bandwidth, application behavior, power supply/consumption, and partial failure. Conventionally, programming in such environment either requires a lot of efforts to build application specific protocols, or to use a middleware that adopts a restricted programming model but still suffers from the loss of performance due to its mismatch with the diversity of the Internet and distributed applications.

In this thesis, we propose to take a unified approach in addressing these issues, by taking advantage of and novelly extending techniques that people learned over decades from various distributed computing area: S-DSM, message passing, remote execution, mobile code and mobile agent, application partitioning, distributed database, World Wide Web, content distribution and multicast. Our belief is that efforts in narrowing the gap between local computing and distributed computing cannot be viewed as a success until middlewares can provide programmers with a natural and rich programming environment that allows efficient implementation of both the middleware itself and applications.

To this end, we propose a middleware system called InterWeave, featuring transparent and efficient data sharing among multiple languages and heterogeneous platforms; scalable and cache coherent hierarchy for data distribution; a combination of data shipping and function shipping enhanced by a shared global space; dynamic application partitioning; version-based relaxed coherence models; 3-level architecture that is able to take advantage of available hardware shared memory at the lowest level, software distributed shared memory across tightly coupled cluster nodes at the second level, and version based coherence in a more loosely coupled environment such as the Internet. In this thesis, we will focus on introducing the first four aspects of InterWeave.