Scheduling Jobs in Multi-Grid Environment Albana Roçi and Reggie Davidrajuh Department of Electronic Engineering and Computer Science
Outline Background Proposed Algorithm Results Conclusion
Multi Grid A grid job is organized as a flow of different activities [1, 2] The grid should be interconnected with some other grids to enable the executions of the tasks [1] Figure 1. Grid System [2]
Activity – Oriented Petri Nets (GPenSim) Model Simplicity Resource management Figure 2. Classical Petri Net Model Figure 3. AOPN Model
Example The first module has 16 tasks and 4 heterogeneous resources. The second module has 17 tasks and 4 heterogeneous resources. Figure 4. First Module Figure 5. Second Module
Proposed Algorithm When an external request comes to the broker, it checks for the availability of the local resource: If the task has been completed The task is in execution mode but the recourse does not contribute in the execution. If the resource is available, but the task has not started the execution because it is waiting for another task under execution to be completed. It checks the time gap between them.
Example Lets suppose that the process is in the 3rd TU. The fifth task from the second module requires first resource from the first module. Figure 6. Graph model of the First Module
Example Figure 7. Depth First Search algorithm Figure 8. Topological sort algorithm
Results Naïve Solution Proposed Algorithm
Conclusion A understandable and efficient algorithm is implemented using GPenSim The time of the execution is minimized The resources are well organized There is no any deadlock during the processing
Thank You!
References [1] M. Silberstein, D. Geiger, A. Schuster, & M. Livny, Scheduling mixed workloads in multi-grids: the grid execution hierarchy. In High Performance Distributed Computing, 2006 15th IEEE International Symposium on (pp. 291-302). IEEE. June 2006. [2] R. Davidrajuh, A New Two-Phase Approach for Petri Net Based Modeling of Scheduling Problems. In Industrial Engineering, Management Science and Applications 2015 (pp. 125-134). Springer, Berlin, Heidelberg. 2015. [3] General Purpose Petri Net Simulator (GPenSIM): http://davidrajuh.net/gpensim/