Using MPI (2nd ed.): portable parallel programming with the message-passing interface
Using MPI (2nd ed.): portable parallel programming with the message-passing interface
Parameter sweeps for exploring GP parameters
GECCO '05 Proceedings of the 7th annual workshop on Genetic and evolutionary computation
Exploring many task computing in scientific workflows
Proceedings of the 2nd Workshop on Many-Task Computing on Grids and Supercomputers
Kestrel: an XMPP-based framework for many task computing applications
Proceedings of the 2nd Workshop on Many-Task Computing on Grids and Supercomputers
Swift: A language for distributed parallel scripting
Parallel Computing
Hi-index | 0.00 |
The complexity and the processing time in scientific experiments based on computational simulation models bring challenges on the conduction of these experiments. Scientific workflows have being adopted on large-scale science. The intense utilization of large volumes of data on these workflows demands parallelism techniques. However, parallelize a workflow requires specific tools and programming skills, which may become a blunder for scientists. To address this issue, this paper proposes Heracles, which is an approach that makes the workflow parallelization into a more transparent task for scientists. Our approach proposes a fault tolerance and dynamic resource management mechanism inspired on P2P techniques. The purpose of Heracles is to execute activities in parallel without asking the scientists to specify the number of nodes involved in the execution and to automatically reschedule failed tasks. In this way, the scientists only need to define the deadline for the workflow. Heracles was evaluated through simulation and showed that it is capable of fulfilling the deadlines of the activities and to recover from failures efficiently. We believe that it may be interesting to integrate Heracles approach on real schedulers to perform deeper evaluations.