Effective scheduling in modern data centers is crucial for optimizing resource allocation, minimizing energy consumption, and ensuring high availability, which ultimately enhances performance, reduces operational costs, and supports scalability in handling diverse workloads. The analysis of recent traces released by Google Borg reveals huge variability in resource demands both in terms of time and resource allocation: for example, the number of cores requested by a job can vary from 1 to several thousands. In this talk, we use a queuing model to investigate the impact of the job scheduling on the data center resource utilization. Furthermore, we present the ‘SMASH’ scheduling, a simple, nearly-FIFO, non-preemptive scheduling discipline that allows for an improvement of the resource utilization with respect to FIFO without posing problems of fairness among the jobs.

About Andrea Marin

Andrea Marin is professor of computer science at the University Ca’ Foscari of Venice, Italy. His research interests include queuing theory and formalisms for performance and reliability evaluation applied to computer and telecommunication systems. Recently, he has studied the problem of scheduling in large data centers and has investigated problems of performance and fairness in blockchain systems.

Este evento se impartirá en inglés