Research And Optimizations On Multiple Storage Media Characteristics Based Storage Energy Efficiency In Datacenters

As the digital universe causes an explosion in data sizes and energy consumption in data centers,storage energy-saving techniques play an important role in achieving carbon neutrality.It is of great significance to reduce grid energy consumption and increase energy efficiency by broadening sources of income and reducing expenditure.However,various storage media with different energy behaviors exist in data center storage systems,making it difficult to improve the storage energy efficiency.The Non-Volatile Memory(NVM)device and green energy provide opportunities for the development of energy-efficient storage systems.But existing solutions are based on publicly available NVM emulators or prototypes,which miss the actual energy behaviors of real NVM devices and are less convincing.In terms of green energy,its intermittent and variable nature makes it difficult for storage systems to efficiently harvest it and reduce traditional energy consumption.In this dissertation,for low grid energy consumption and high green energy utilization,we present energy-efficient solutions for NVM,SSD,and HDD storage systems in datacenters.In the first section of this dissertation,we investigate the energy-efficient designs for all-NVM storage systems using real NVM devices.NVM devices offer comparable performance with DRAM while providing larger capacities and data persistence with lower energy consumption,and they can smooth the performance gap between memory and storage devices.However,all-NVM storage systems in practice suffer from non-trivial energy consumption and high hardware temperatures.To this end,we propose an energyefficient request gathering approach,named Sprint-NVM.It consists of energy-aware caching,prefetching,and data syncing designs that significantly improve the energy efficiency.Sprint-NVM serves the majority of requests with high-power primary NVMs,and the requests of the remaining NVMs will be converted to compact requests to improve the energy efficiency.Data exchanges between the primary and remaining NVMs are in a compact and parallelized way,allowing the remaining NVMs to stay in low-power mode for a longer period.Experiment results show that Sprint-NVM saves up to 26% energy with negligible performance degradation when compared to energy-unaware NVM approaches.In the second part of this dissertation,we propose to match the peaks and valleys between green energy and storage workloads for NVM-SSD tiered storage systems.Taking advantage of green energy is becoming increasingly popular,however,when it comes to the storage I/Os,the irregular fluctuations of green energy lead to the mismatch of peaks and valleys between green energy and storage workloads.To address this issue,we propose a tiering-cooperative energy-efficient scheme,called Mix Save.It efficiently tiers NVM and SSD and replicates data among them based on their performance and energy characteristics,achieving both high performance and large capacity.Mix Save customizes the energy-saving strategies for each layer and tunes them cooperatively between layers,allowing Mix Save to reduce overall energy usage.The workload-driven strategy is used by the NVM layer,which is used for caching,to ensure the best performance,and the number of active NVM devices depends on the storage workload.Meanwhile,the SSD layer employs a green energyoriented strategy to conserve traditional energy,and the number of active SSD devices will be proportional to the amount of green energy supplied.Evaluation results show that,when compared to energy-save-unaware approaches,Mix Save degrades performance by less than4%and saves traditional energy by 73%～80%and 55%～61%under light and heavy workloads,respectively.In the third part of this dissertation,we propose to improve the utilization of green energy and save grid energy for SSD-HDD hybrid storage systems.Because storage systems adjust the number of active storage devices based on storage workloads or green energy supplies,delayed I/O scheduling degrades performance and wastes green energy.We propose Pre Match to address the delayed scheduling issue for the SSD-HDD hybrid storage system.First,energy-efficient SSD servers are used to host hot data,while cost-effective HDD servers with large capacities are used to store cold data to reduce storage costs,and some HDD servers can be turned off to save energy.To make decisions in advance,we use an LSTM neural network to forecast workload and green energy.A balance-scheduling energy scheduling scheme is proposed to make active devices proportional to the dominant one of workload and green energy.When compared to the Workload-Driven Scheme(WDS),evaluation results show that Pre Match halves grid energy consumption and has higher green energy utilization.Pre Match also achieves the same performance as WDS.