Homework Four

Assignment:

1. Read "Characterization of Scientific Workloads on Systems with Multi-Core Processors", by S.R. Alam, R.F. Barrett, J.A. Kuehn, P.C. Roth, and J.S. Vetter (http://ft.ornl.gov/~vetter/pubs/2006-10_iiswc06_multicore.pdf). Seek other information on the performance of these processors particularly on floating-point and double precision computations, and particularly from sources other than the manufacturers. Using this, wtite a short (two to four pages including illustrations) description of dual and quad core Intel and AMD processors and their performance. Pay particular attention to potential advantages and disadvantages of multi-core processors, and evaluations of their performance on different applications. Include appropriate references to all sources used. Hint: Try Tom's Hardware page for performance information (http://www.tomshardware.com/)
2. Perform an operation count on the three versions of Sieve of Eratosthenes - the cache unfriendly, cache friendly and the multi-threaded cache friendly versions. Describe the amount of overhead added by each of the cache friendly and the multi-threaded cache friendly versions.

   Compile and run the test programs in Akhter Chapter 7 on dune and strider. Modify the multi-threaded cache friendly versions to run on 1, 2, 4 and 8 threads on dune, and 1, 2 and 4 threads on strider and comment on the results.

3. Examine the effects of modifying the order of the loops that traverse rows and columns in the matrix multiplication routine omp_mm.c and also varying the chunck size. Run on 1, 2, 4 and 8 threads on dune, and 1, 2 and 4 threads on strider and comment on the results.

   Read Dr Bramley's