Research On Computing Resources And Power Sources Collaborative Management Approaches In Green Data Centers

With the development of big data,AI,and cloud computing technology,data center business has been growing explosively.At the same time,the scale of data center is expanding rapidly and the cost of building and maintaining data center every year is huge.The high energy consumption of data centers not only brings huge electricity costs,but also brings enormous pressure to the environment.Therefore,academia and industry are committed to introducing clean energy into traditional data centers with grid as the main power source and building green data centers to reduce operating costs and greenhouse gas emissions,such as renewable energy(wind and solar energy).However,the existing green data center design still faces an important challenge from the power mismatch between server power demand and renewable power supply.Therefore,there is an urgent need of efficient scheduling policies to match the server power demand and the renewable power supply.To deal with the power mismatch problem,many research works have been proposed from the aspects of the underlying server and renewable power managements.However,these efforts have not fully exploited the heterogeneous characteristics of hardware in current data centers and the potential of dynamic scheduling of processors,i.e.efficient coordination of workload,server hardware and renewable power characteristics,so as to optimize the overall energy efficiency and performance of green data centers.This paper will conduct in-depth research from three following specific aspects.To achieve a collaborative management for heterogenous servers and reduce the power consumption of data centers,we propose a runtime power management system based on a latency-aware feedback control mechanism for heterogeneous servers.Heterogeneous servers have already existed and been leveraged to improve energy efficiency data centers in recent years.However,running latency-critical web-services on heterogeneous servers is still challenging because the overheads of task transition between such servers have great negative impacts on performance.Montgolfier consolidates wimpy and brawny servers into composite nodes serving latency-critical applications to improve overall energy efficiency while ensuring QoS.In order to eliminate the additional overhead caused by task migration,Montgolfier uses dynamic load forecasting method to predict the load intensity of data center in short term,so as to accurately determine the time of load switching between servers.Based on the exponential weighted moving average method,the prediction method is optimized by adding a negative weight system to the original single weight coefficient to improve the accuracy of peak load forecasting.At the same time,the optimization selection method of processor power allocation,such as DVFS technology,is designed to cope with the fluctuation of load requests.The method uses a double threshold control mechanism to fine-grained control the real-time response delay within a safe range.Finally,Montgolfier-Mix strategy is designed to improve the utilization of server resources.This strategy turns on two heterogeneous servers of heterogeneous computing nodes and achieves the goal of running batch processing and delay sensitive tasks simultaneously.Our evaluation results show that Montgolfier reduces energy consumption by up to 34.9% without violating any QoS constraints.Compared with the static wrokload scheduling strategy,Montgolfier-Mix can improve the performance of batch tasks up to 53%.To achieve a collaborative management between renewable power and heterogeneous servers and propose an efficient power allocation,we propose a dynamic power allocation framework called GreenHetero,which enables adaptive power allocation among heterogeneous servers in green datacenters to achieve the best performance when the renewable power varies.Considering the system and hardware upgrade of the data center,such as server migration,replacement of failed server components,computing and storage capacity expansion of the server,the existing data center presents more heterogeneity.However,for green data centers,the existing research work has not proposed effective power distribution schemes for sensing the heterogeneity of the underlying data center,that is,how to maximize the overall performance of the data center under the limited renewable power supply.GreenHetero is a dynamic power allocation strategy,which uses Holt exponential smoothing power prediction algorithm to help power source selectors select appropriate power source from renewable power,batteries and grid power.The prediction algorithm can capture the trend information of current monitoring data,and accurately predict the renewbale power supply and server power demand.In order to match the diversity of workload and platform,GreenHetero creates a database that includes performance-power relationship for all server configurations and workloads,and maintain and updated it in real time through the monitoring mechanism and the lightweight sampling method.Based on the database information and power forecasting mechanism,GreenHetero finally determines the power allocation ratio of all heterogeneous servers,and implements power selection and power allocation decisions through the power enforcer module.The experimental results show that GreenHetero can achieve a performance improvement of up to 2.2 times under the same power supply conditions compared with homogeneous strategy.To achieve a collaborative management between Computational Sprinting and renewable power and boost the computing performance of data centers,we propose GreenSprint,a renewable energy driven approach that enables a data center to boost its computing performance efficiently by conducting computational sprinting under the intermittent and time-varying nature of renewable energy supply.Computational Sprinting has proven to be an effective way to boost the computing performance for bursty workloads.However,extra power available for sprinting is constrained by existing power distribution infrastructures.GreenSprint is a renewable energy driven approach to operate sprinting activities in green data centers.We also present three basic strategies designed to determine the core count and frequency level for sprinting based on current power supply.Furthermore,we propose a Hybrid strategy that combines reinforcement learning to dynamically determine the optimal server setting,targeting at both the power provision safety and the quality of service.In consideration of practical cooling conditions,we also present a thermal-aware sprinting strategy Hybrid-T.Finally,we build an experimental prototype to evaluate GreenSprint on a cluster of 10 servers with a simulated solar power generator.The results show that renewable energy by itself can sustain different duration lengths of sprinting when its supply is sufficient and can improve performance by up to 4.8x for representative interactive applications.The above three aspects of research have realized the efficient collaborative scheduling management of computing resources and power resources in green data center,and provided necessary theoretical and key technology support for the future construction of efficient green data center.