Distributed systems

Distributed systems consist of a large number of interconnected computers that provide a high computational power. They are at the base of cloud computing and often are required to perform analysis of big data by using distributed algorithms such as MapReduce.

Quantitative analysis of distributed systems address many crucial aspects among which:

  1. How do we distribute the workload among the computational units in order to maximize the  system’s performance?
  2. How do we relocate the work among the computational units in the case of  highly unbalanced conditions?
  3. How do we handle the strict network requirements of data centers?

ACADIA research group on performance evaluation has addressed these problems both from practical and theoretical point of view. More specifically, it has developed a new algorithm to assign the computational resources in split-merge computations (e.g., MapReduce), and a model that allows the tuning of a distributed system which migrates the jobs in order to improve its responsiveness.

These works received the  best contribution to the conference award at the International Conference Valuetools 2017.

Selected publications

  1. A. Marin, S. Rossi, M. Sottana: Biased Processor Sharing in Fork-Join queues
  2. A. Marin, S. Balsamo, J.-M. Fourneau: LB-networks: A model for dynamic load balancing in queueing networks. Perform. Eval. Volume 115, pp. 38-53, 2017
  3. S. Rossi, A. Marin: Fair workload distribution for multi-server systems with pulling strategies. Perform. Eval. Volume 113, pp. 26-41 2017