Green Data Center
Brief Introduction
Today’s giant data centers are power hungry. Data center energy saving not only controls the operational cost, but also benefits the environments. IT equipments dominate the energy consumption in modern data centers and it is increasingly important to reduce the energy consumed by the network part of IT equipments. This project aims at designing energy efficient flow scheduling mechanism to reduce the network energy footprint of distributed computation in container-sized data centers. A central controller periodically schedules the flows to accommodate flow dynamics, network dynamics and inaccurate estimation. Using Fat-Tree as a representative data center network architecture, we exploit the topological characteristic to develop an online flow scheduling algorithm. We evaluate the topology-aware flow scheduling by real MapReduce workloads from Terasort and Wordcount applications. Simulation results show that the algorithm can save 15%~40% network energy compared with the ECMP flow scheduling. It also outperforms the classical optimization algorithms such as simulated annealing and particle swarm optimization, in terms of both the network energy saving and the scheduling speed.
Today’s giant data centers are power hungry. Data center energy saving not only controls the operational cost, but also benefits the environments. IT equipments dominate the energy consumption in modern data centers and it is increasingly important to reduce the energy consumed by the network part of IT equipments. This project aims at designing energy efficient flow scheduling mechanism to reduce the network energy footprint of distributed computation in container-sized data centers. A central controller periodically schedules the flows to accommodate flow dynamics, network dynamics and inaccurate estimation. Using Fat-Tree as a representative data center network architecture, we exploit the topological characteristic to develop an online flow scheduling algorithm. We evaluate the topology-aware flow scheduling by real MapReduce workloads from Terasort and Wordcount applications. Simulation results show that the algorithm can save 15%~40% network energy compared with the ECMP flow scheduling. It also outperforms the classical optimization algorithms such as simulated annealing and particle swarm optimization, in terms of both the network energy saving and the scheduling speed.
My Contribution
I studied the topology in data center and heuristic algorithms including simulated annealing and particle swarm. I also implemented our own topology-aware flow scheduling algorithm simulation using C++ and made comparisons with other heuristic algorithms.
I studied the topology in data center and heuristic algorithms including simulated annealing and particle swarm. I also implemented our own topology-aware flow scheduling algorithm simulation using C++ and made comparisons with other heuristic algorithms.