ReRack: power simulation for data centers with renewable energy generation
ACM SIGMETRICS Performance Evaluation Review
| Hi-index | 0.00 |
This work contends that three constraints need to be addressed in the context of power-aware real-time systems: energy, time and task rewards/values. These issues are studied for two types of systems. First, embedded systems running applications that will include temporal requirements (e.g., audio and video). Second, servers and server clusters that have timing constraints and Quality of Service (QoS) requirements implied by the application being executed (e.g., signal processing, audio/video streams, webpages). Furthermore, many future real-time systems will rely on different software versions to achieve a variety of QoS-aware tradeoffs, each with different rewards, time and energy requirements. For hard real-time systems, solutions are proposed that maximize the system reward/profit without exceeding the deadlines and without depleting the energy budget (in portable systems the energy budget is determined by the battery charge, while in server farms it is dependent on the server architecture and heat/cooling constraints). Both continuous and discrete reward and power models are studied, and the reward/energy analysis is extended with multiple task versions, optional/mandatory tasks and long-term reward maximization policies. For soft real-time systems, the reward model is relaxed into a QoS constraint, and stochastic schemes are first presented for power management of systems with unpredictable workloads. Then, load distribution and power management policies are addressed in the context of servers and homogeneous server farms. Finally, the work is extended with QoS-aware local and global policies for the general case of heterogeneous systems.