Energy Efficiency Optimization Mechanisms For Data Centers With Multi-source Power Supply

With the wide deployment of cloud computing datacenters, the problems of tremendous power consumption, high operational cost and serious environmental pollution have become increasingly prominent. Renewable energies, smart grids and efficient energy storage devices have brought new opportunities for energy-efficiency optimization of cloud datacenters. However, renewable energies are unstable and intermittent, and smart grids offer multiple markets and time-varying prices, which incur many new challenges to realize efficient and reliable green datacenters. Optimizing energy-efficiency under renewable energy and smart grids environment, provides significant advantages to construct environment-friendly, resource-efficiency and adaptive large-scale datacenters.The research on energy-efficiency optimization of multiple power source datacenters is emerging in recent years. However, there are still many issues that need to be further studied. First, existing research works either assume a priori knowledge of workloads demand, or require a substantial amount of statistics of the system dynamics, obtained based on excessive computational complexity or different forecast techniques. Traditional approaches to construct optimal control policies involve the use of Markov Decision Theory and Dynamic Programming. These techniques suffer from the "curse of dimensionality" where the complexity of computing the optimal strategy grows with the system size. Second, current works neglect the interaction among energy production/purchase, energy storage and demand management from the prospective of datacenter operators. Third, with the presence of fluctuating workloads in datacenters, the lifetime and reliability of servers under dynamic power-aware consolidation could be adversely impacted by repeated on-off thermal cycles, ware-and-tear and temperature rise. Software also exhibits an increasing failure rate under heavy load over time. Fourth, there is a lack of study in the existing literature on quantifying the role and cost tradeoff of rightsizing the battery for energy store in a cloud datacenter. Finally, it remains a significant challenge how to design online, lightweight, decentralized and multi-regional energy efficiency-aware request routing, load scheduling and energy management algorithm for geo-distributed datacenters.First, we propose novel adaptive control strategies of energy storage devices under different pricing schemes. The goal is to obtain a thorough understanding on the tradeoffs between battery’s contribution in energy storage and its provisioning cost, in order to gain useful insights for battery rightsizing in cloud datacenters with arbitrary energy demand and time-varying renewable energy generation. Under each pricing scheme, we formulate the problem as a stochastic optimization model that minimizes the long-run operational cost of a datacenter, and derive adaptive, optimal control policies that guide power purchase from the smart grid and battery charge/discharge efficiently. Further, we theoretical analyze that under what condition the investment in energy storage is beneficial, and we quantify the degree to which battery can help mitigate the impacts of variable renewable energy generation, as well as the needed capacity of battery.Second, we propose an online control policy of demand scheduling and collaborative optimization of multiple power sources for datacenters, called SmartDPSS. Without requiring a priori knowledge of system statistics, SmartDPSS allows CSPs to make online decisions. Based on the two-stage Lyapunov optimization techniques, SmartDPSS makes two-timesacle decisions:to decide the amount of energy to purchase from the long-term-ahead grid market in intervals of longer periods of time, to tackle demand dynamics and energy price fluctuations in the future intervals, and also to decide the amount of energy to purchase from the real-time grid market, as well as the amount of energy to store into or discharge from the UPS battery, in smaller time scales. The online decisions are set to best utilize the available renewable energy produced over time and the periods with lower electricity prices from the grid market, in order to minimize the operational cost in the long run of the datacenter. Thoroughly analysis and simulations verify the optimality, robustness and scalability achieved by SmartDPSS.Third, we propose a novel Reliability-Aware server Consolidation stratEgy, named RACE. The focus is on the characterization and analysis of this problem as a multi-objective optimization, by developing an utility model that unifies multiple constraints on performance SLAs, reliability factors, and energy costs in a holistic manner. An improved grouping genetic algorithm is proposed to search the global optimal solution, which takes advantage of a collection of reliability-aware resource buffering, and virtual machines-to-servers re-mapping heuristics for generating good initial solutions and improving the convergence rate. Extensive simulations are conducted to validate the effectiveness, scalability and overhead of RACE.At last, we propose an online green load-balancing policy for geographical distributed cloud datacenters, called GeoGreen. Based on two-timescale Lyapunov optimization techniques, without the need to predict future system information, GeoGreen can balance workload across geo-distributed datacenters and operate multiple power sources of each datacenter in a complementary manner to minimize the long-term operational cost. By adjusting the cost-delay parameter, cost-reliability parameter and operation frequency parameter, GeoGreen enables cloud service providers to gain desired tradeoff between operational cost and service latency.In summary, our proposed mechanisms about workload online scheduling and collaborative management of multiple power sources from different levels of cloud datacenters, can reduce the operational cost significantly and improve system energy-efficiency dramatically.