Simulation and Modelling of Task Migration in Distributed Systems Using SimGrid

Developing protocols and algorithms in the field of distributed computing necessitates achieving comparable results from a large number of back-to-back experiments on real distributed systems. However, while multithreading cannot provide sufficient performance for a real distributed system, using real systems is usually prohibitively expensive and necessitates highly skilled resources and power management. Simulation is used as a solution to address these issues in that it does not only allow for reliable results at a low cost and in a reasonable amount of time but also allows for the exploration of a wide range of platforms and scenarios. Churn is the most concerning issue when modelling and simulating a distributed network. It refers to the unpredictability of a large number of arriving and departing devices in a short period of time. In response to Churn, the server initiates Task Migration, which entails relocating the remaining jobs to another device in the same network. In this paper, we use the simulation framework SimGrid to run two experiments on task migration in distributed systems. For the same task, we use a different network size in each experiment. To ensure the network’s heterogeneity, we select the dataset GWA-T-13 Materna, which contains performance metrics described as trace files of over 1500 VMs from the distributed Materna Data Centres in Dortmund, Germany. In both experiments, the simulation results show how the distributed system responds to Churn in a way that reflects in the execution time and power consumption of those devices that initiate task migration.