Mission

One way of achieving significant improvements in the speed of computers is to use many CPUs to cooperate in solving a single problem. Unfortunately programming the cooperation of multiple computers is very difficult and expensive. So much so that it is seldom used in practice. In particular, it is usually necessary to completely rewrite programs that already work on single computers to work on multiple ones.

This research proposes a new way of coordinating multiple computers using an algorithm called Time Warp which will be embodied in a new CPU called the WarpEngine. The coordination is achieved by the hardware of the computer rather than by the programmer rewriting the program. The research will seek to find out if the technique is feasible, how hard the resulting machine will be to build and what performance can be expected on real programs. This will be done by doing simulations of the design rather than actually building the CPU. If the research is successful it will provide a way of easily and cheaply speeding up a wide range of existing computer programs.

This project is being supported by a $50,000 Marsden grant.