Group Home

Data Centers Home






This work is in collaboration with Prof. Yuanyuan Zhou.

Power management has become critical for high-end data centers. In these systems, significant power is consumed by multiple components (e.g., processors, memory, and disks). Further, these systems also demand predictable performance. We have explored performance-guaranteed energy management algorithms for adaptive processors, memory, and disks. Previous algorithms require painstaking, application-dependent manual tuning, can incur arbitrary performance degradation, and treat individual components in isolation without considering their (sometimes conflicting) interactions. We have (1) developed a technique to enforce performance guarantees while adapting for energy, (2) proposed adaptation control algorithms that require no hand-tuning, guarantee performance, and save significant energy, and (3) developed a technique that jointly minimizes processor-memory energy. Our techniques make a significant step towards practically usable system-wide energy management algorithms. We are currently studying our algorithms for commercial database workloads.