Computer Science 595N, Fall 2010:
Tools for High-Performance Computing with Big Graphs

John R. Gilbert

Tue 1:00-2:30, Harold Frank Hall 1152

Computation with combinatorial and discrete structures -- graphs, networks, strings, partial orders, etc. -- has become ubiquitous in many areas of data analysis and scientific modeling. However, the field of high-performance combinatorial computing is in its infancy, and computations with discrete structures do not scale well on multicore and future manycore processors.

By contrast, in the mature field of numerical high-performance computing, programmers possess standard algorithmic primitives, high-performance software libraries, powerful rapid-prototyping tools, and a deep understanding of effective mappings of problems to high-performance computer architectures. A key challenge of the manycore revolution is to replicate these achievements for combinatorial computing.

In this seminar class, students will read and present papers on different research groups' approaches to tools for high-performance combinatorial computing, and in particular for manipulating large graphs and networks using parallel computers. I'll have a list of suggested papers to read at the first meeting on Tuesday, September 28.

There will also be an opportunity for students who wish to do programming projects, either for separate research credit or for presentation in CS 595. This is in connection with with the Combinatorial Scientific Computing Lab's work on our "Graph BLAS" library. Here are a writeup and a talk with some background on the Graph BLAS.

Possible topics and papers to present

Schedule of Talks: