Towards the Prediction of the Performance and Energy Efficiency of Distributed Data Management Systems
- Raik Niemann
- ICPE Companion
Nowadays developers and end users of data management systems are challenged with the reduction of the "energy consumption footprint" of existing implementations and configurations. In other words, the energy efficiency has to be optimized, either by increasing the performance or by consuming less resources. In fact, there is a big number of factors that influence the performance and energy efficiency of a particular data management system. For example, the replacement of hardware components or the surrounding operating system can have a significant impact. Both developers and end users put much effort into finding performance "bottlenecks", better hardware resource utilization and configurations. Besides, when it comes to a scale-out scenario, end users often face the situation to find a hardware configuration that offers both a reasonable performance and energy consumption, i.e. a resource planning. This paper proposes a new approach to evaluate the performance of a data management system and the impact on the energy efficiency with the goal to optimize it. The approach introduces a Queued Petri Nets model, whose simulation runs are intended to drastically reduce the investments, both in time and hardware, compared to traditional ways, for example regression and compatibility tests. The model's prediction in terms of performance and energy efficiency were evaluated and compared to the actual experimental results. On average the predicted and experimental results (response time and energy efficiency) differ by 24 percent.