Simulating a Multi-Layered Grid Middleware

Quentin Guilloteau (1, 2, 3)

1 UGA - Université Grenoble Alpes
2 DATAMOVE - Data Aware Large Scale Computing
3 CTRL-A - Control for Autonomic computing systems

The study of grid or cluster middlewares is complex, and experiments on such systems are costly. This cost can come from the number of resources required to deploy realistic experiments, or the time to replay a significative workload. Simulation techniques can help reduce such cost to almost none and help perform preliminary study at low cost, but they might also degrade the realism of the results. In this paper, we consider the implementation in simulation of the CiGri grid middleware, in Batsim, a batch sheduler simulator. We are particularly interested in the impact of simulation on signals of interest and their dynamics compared to the real system in the optic of using this simulator to accelerate the first steps of future studies.

Domains: Distributed, Parallel, and Cluster Computing [cs.DC]

Quentin Guilloteau. Simulating a Multi-Layered Grid Middleware. 2023. ⟨hal-04101015⟩