CIS5930final08 - 6. Why is the speedup for master-worker...

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
HPC Final Exam - Fall 2007 Name: (Please print) Put the answers on these sheets. Use additional sheets when necessary. You can collect 100 points in total for this exam. 1. What does SPMD stand for? (5 points) 2. What is the difference between task partitioning and data partitioning? (5 points) 3. Explain the blocking, locally-blocking, and non-blocking send operations, e.g. which send operations suspend, and which allow immediate reuse of the buffer? What are the three corresponding MPI send functions? (10 points) 4. Give an explanation of functionality of the broadcast and scatter collective communications. Both operations may use a tree-based message passing pattern. Explain the tree-based mes- saging (e.g. using an example) and give analytical estimates of the communication time t comm for broadcast and scatter of n array elements. (15 points) 5. The Barnes Hut algorithm relies on a simplifying assumption to avoid calculating all of the n ( n - 1) interactions between n particles. What is this simplifying assumption? (10 points)
Background image of page 1
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: 6. Why is the speedup for master-worker parallel implementations of low-computation tasks (such as geometric transformations on images) very poor? (10 points) 7. Explain the dynamic load balancing technique referred to as the fully distributed work pool. (10 points) 8. Give the three steps of the divide and conquer parallelization strategy. (10 points) 9. Give an outline of the fully recursive divide and conquer parallel matrix multiply algo-rithm. (10 points) 10. Consider square matrix multiply with the fully recursive divide and conquer parallel matrix multiply algorithm. We divide the n n in four blocks, thus halving the matrix dimension n in each recursive step. The amount of work to combine the results per step is n 2 (four matrix additions in parallel, each addition taking n 2 operations). Give a formula for the total execution time t comp as a function of n (you may assume that n = 2 k ). (15 points) 1...
View Full Document

Ask a homework question - tutors are online