Use of Bio-Inspired Algorithms for the Efficient Management of Geo-distributed Data Centers

The talk will introduce the state-of-the-art regarding the cost- and energy- efficiency of Cloud platforms and data centers, and will highlight the necessity of improving the computational efficiency of the IT components (servers, VMs, storage, etc.), while the physical efficiency (cooling, power distribution, etc.) has improved a lot recently. The talk will describe innovative techniques used to optimize the allocation and scheduling of virtual machines on the hosts of a data center. These techniques are based on recent results in the fields of bio-inspired distributed algorithms and swarm intelligence, which allow the efficient handling of a problem that would otherwise be intractable (NP-hard) if tackled with a classical centralized approach. Then, the talk will illustrate how a hierarchical approach can help to manage a multi-data center environment, where the adoption of the "follow the moon" paradigm - i.e., move the workload where it can be executed more efficiently or with lower energy consumption - can be profitably exploited to achieve numerous business goals, while preserving the autonomy of single data centers. Mathematical models and formalisms will be presented, with concrete examples, to support and give a foundation to the experimental results achieved on real data centers.