|
| ||
course participants course announcements about this wiki questionnaires and assignments slides of presentations course schedule related resources  |
The thesis focused on a method to schedule large workflows onto distributed, heterogeneous Grids. However, the author not only focused on efficient algorithms to map workflows onto Grids, but he also created, implemented, and tested an actual tool to schedule large workflows (in the domains of bio-imaging and astronomy) onto a Grid. A main idea in the thesis was that if scheduling algorithms focused on “planning ahead” by utilizing various performance models then those algorithms would hopefully reduce the turn around time for a workflow on a Grid. The idea behind planning ahead is that all scheduling decisions are made before any portion of the workflow is executed, as opposed to performing the scheduling as portions of the workflow are executed, which is how current dynamic scheduling algorithms function. A key component to this method of “planning ahead” is creating a performance model of the situation so all decisions can be made ahead of time. The author devised performance models for communication performance as well as the component performance. He then provided a heuristic approach that attempts to map the workflow onto the Grid, and he provides a guarantee that they run in polynomial time, but aren’t guaranteed to provide an optimal mapping. Overall I felt that this thesis was deserving of a Ph.D. The author thoroughly covered the appropriate background material, his approach along with related work, and an analysis of the drawbacks of his approach, and how they might be addressed. In addition to his work he also considered the scalability of his approach on large Grids. Lastly, he ended with a section detailing possible future work over the short term, medium term and long term. I felt that he provided a genuine contribution to computer science by addressing an important area, namely the scheduling of resources, in the emerging area of Grid computing. I also feel that his creation of a tool to implement his ideas and test them in a real situation provided valuable information regarding the usefulness and application of his creation, as opposed to just creating theoretical models and algorithms. His tests showed an improvement of turn-around time for workflows when compared against previous dynamic scheduling algorithms and tools. The only major complaint I have against his work is he doesn’t seem to have any cases where a dynamic scheduling algorithm and tool outperform his algorithm and scheduling tool – though it may be possible that his outperforms all other methods in as many cases as it is reasonable to test I don’t feel that he clearly addressed it (by providing an example where previous methods still work better than his proposed method, or by claiming that his works better than all other methods). Last modified 2 December 2007 at 11:07 am by paul.marshall |