home · mobile · calendar · colloquia · 1995-1996 · 

Colloquium - Tanenbaum

Parallel Programming on a Distributed System
Vrije Universiteit

There are two ways to build a parallel computer system: as a multiprocessor (i.e., with shared memory) or as a multicomputer (i.e., no shared memory). The former is easy to program but hard to build. The latter is easy to build but hard to program. In this talk I will describe a hybrid system we have designed and implemented that combines the advantages of both approaches.

Our model is based on simulating a restricted form of object-based shared memory on a multicomputer. Shared objects are replicated in the memory of each machine that uses them. Read operations are done locally, without requiring network traffic. Writes can be done using a reliable broadcast algorithm if the hardware supports broadcasting; otherwise, a point-to-point protocol is used. The semantics presented to the application are that of sequentially consistent operations on objects.

We have built a prototype based on the Amoeba distributed operating system, and implemented a parallel programming language, Orca, on top of it. The talk will discuss Amoeba, Orca, some initial applications, and the performance of the system.

Refreshments will be served prior to the talk.

Department of Computer Science
University of Colorado Boulder
Boulder, CO 80309-0430 USA
May 5, 2012 (14:13)