Project for Graduate Course on Computer Architecture:

Experiment: To determine if the QSM is a suitable model for designing algorithms to run on a network of workstations.

The QSM is a shared-memory parallel computing model together with a bulk-synchronous programming model [GMR97]. Results have been obtained to show that it is as effective as the message-passing BSP model (which is similar in many ways to the recent LogP model), while being simpler and providing a shared-memory programming environment. Experimental results also suggest that the QSM is a useful model on which to develop parallel algorithms for the Cray C90, Cray J90 and the MasPar MP-1 -- all of which are tightly-coupled parallel machines (see [BGMZ95,GMR96] -- these experiments implemented algorithms developed for the QRQW PRAM, which is the precursor to the QSM [GMR94a,GMR96]).

In this project you will develop experiments to evaluate to what extent the QSM models computations on the LESS network of workstations. Here is one suggestion for an experiment.

Run algorithms designed for the QSM on the LESS network and compare their performance with the best algorithms designed for the same problems on other high-level parallel models. Also, compare their performance to an implementation you design that is tailored to the LESS network. Evaluate the benefits of the QSM with respect to the trade-off between simplicity, accuracy and wide applicability. You may need to measure L and g (and also perhaps o) for the LESS network in order to interpret the results you obtain.

This is just one suggestion for a project relating to the QSM. One could think of other experiments that would evaluate other features of the QSM.

REFERENCES:

[BGMZ95] G.E. Blelloch, P.B. Gibbons, Y. Matias, M. Zagha, ``Accounting for memory bank contention and delay in high-bandwidth multiprocessors,'' Proc. ACM SPAA, 1995.

[GMR94] P. Gibbons, Y. Matias, V. Ramachandran, ``The QRQW PRAM: Accounting for contention in parallel algorithms,'' SIAM J. Comput, to appear; preliminary version in Proc. ACM-SIAM SODA, 1994, pp. 638-648.

[GMR96] P. Gibbons, Y. Matias, V. Ramachandran, ``Efficient low-contention parallel algorithms,'' J. Computer and System Sciences (Special Issue on papers from ACM SPAA'94), vol. 53, No. 3, 1996, pp. 395-416.

[GMR97] P. Gibbons, Y. Matias, V. Ramachandran, ``Can a shared-memory model serve as a bridging model for parallel computation?'' Proc. ACM SPAA, June 1997, pp. 72-83.

Papers [GMR94], [GMR96], and [GMR97] are available on the web at http://www.cs.utexas.edu/users/vlr/pub.html.