Publication details

Agent-based User Modeling in Job Scheduling Simulations

Authors

TÓTH Šimon KLUSÁČEK Dalibor

Type Article in Proceedings
Conference Cracow Grid Workshop
MU Faculty or unit

Faculty of Informatics

Citation
Field Informatics
Keywords Grid; Simulation; Job scheduling; Workload
Description In this paper we present an agent-based user model extension for the Alea simulator. This model is a necessary prerequisite for the support of dynamic workload adaptation. Also, even when used with static workload traces, it enables greater flexibility and control over how and what jobs are submitted into the queues of the simulated scheduler. With the inclusion of the agent-based user model into the Alea simulator, we were able to implement support for dynamic workload generation, which in conjunction with the agent model greatly improved realism of our simulations. We are now able to properly evaluate changes in users' workloads and test scheduling approaches that rely on the dynamic user behavior. Our future work will concentrate on formulating a complete evaluation methodology based on the agent-based model and dynamic workload generation.
Related projects: