Autonomic and Energy-Efficient Management of Large-Scale Virtualized Data Centers

摘要: Large-scale virtualized data centers require cloud providers to implement scalable, autonomic, and energy-efficient management systems. To address these challenges this thesis provides four main contributions. The first one proposes Snooze, a novel Infrastructure-as-a-Service (IaaS) system, which is designed scale across many thousands of servers virtual machines (VMs) while being easy configure, highly available, energy efficient. For scalability, Snooze performs distributed VM based on hierarchical architecture. support ease configuration high availability implements self-configuring self-healing features. Finally, for efficiency, integrates holistic approach via resource (i.e. CPU, memory, network) utilization monitoring, underload/overload detection mitigation, consolidation (by implementing modified version the Sercon algorithm), power transition idle into saving mode. A modular prototype was developed extensively evaluated Grid'5000 testbed using realistic applications. Results show that: (i) does not impact submission time; (ii) fault tolerance mechanisms do application performance (iii) system scales well with an increasing number resources thus making it suitable managing large-scale centers. We also that able dynamically center consumption its allowing conserve substantial amounts only limited performance. open-source software under GPLv2 license. second contribution placement algorithm Ant Colony Optimization (ACO) meta-heuristic. ACO interesting due polynomial worst-case time complexity, close optimal solutions parallelization. Simulation results scalability current implementation smaller VMs, outperforms First-Fit Decreasing greedy in terms required computes solutions. In order enable scalable consolidation, makes two further contributions: ACO-based algorithm; fully decentralized unstructured peer-to-peer network. key idea apply small, randomly formed neighbourhoods servers. our by emulation state-of-the-art algorithms V-MAN) algorithm. be as achieve obtained executing centralized