Special Topics: The BSP Model for
Portable & Efficient Parallel Computing

---

COP 5937, Section 1

Parallel computing has long been heralded as the wave of the future. Decades of experience, however, have demonstrated that creating efficient and portable parallel software is a difficult task, largely due to the diversity of current parallel computers and parallel computer models. Some use shared-memory techniques, others use message passing. Some take advantage of network locality, others do not. As a result of this chaotic situation, parallel application code is usually computer specific - to run the application on a different parallel computer generally requires a fundamental restructuring of the solution method.

In order to make parallel computing more manageable, therefore, it would be desirable to have a unifying model for parallel computing. An acceptable unifying model should have several properties. First, it should be simple enough to allow programmers to reason about and write parallel code. Second, it should be close enough to physical reality that computer architects can design efficient hardware. Third, it should be mathematically tractable to aid in the development and analysis of algorithms. The Bulk-Synchronous Parallel (BSP) model, proposed by Leslie Valiant, is a promising candidate for the unifying model.

The purpose of this course is to familiarize the student with the theory behind the BSP model and its practical consequences. The course will consist of a series of group discussions - students will be expected to present their own ideas and opinions. Furthermore, this is a fruitful research area. Students seeking cutting-edge research projects that are of both practical and theoretical significance are strongly encouraged to enroll.

This course was first offered in Fall 1995 and was very well-received by students.