An Energy-saving Storage System Of Astronomical Science Data With Software And Hardware Co-Design

Kunlun Station in Antarctica is excellent for astronomical observation and China has been deploying telescopes in this area.However,observation data cannot be delivered for analysis in time due to the poor network conditions there.Therefore,it is necessary to establish a data center there.Storage systems generally consumes much energy.Therefore,the limited power supply of Kunlun Station becomes a bottleneck for its data center.Many general-purpose energy-efficient strategies are designed to be applied in common data centers,which is quite different from the disk array at Kunlun Station where extreme restrictions would influence the effect of these solutions.Besides,maintaining the reliability is as important as saving energy because most of the time,nobody is there to solve the disk failure problems.This thesis investigates the related work about energy-saving disk storage and summarizes the applicable scenarios and shortcomings of hardware-based energy-saving technologies such as RAID and variable-speed disks and software-based methods such as data caching,prefetching,copying and migration.Based on the needs of the storage system at Kunlun Station,this thesis proposes a data-aware energy-saving storage management strategy,named DAES,for astronomical observation whose purpose is to reduce energy consumption while mitigating the loss of the reliability of disks.A metric named hit index is designed for each disk from the perspective of astronomy to manage the power state of disks more accurately.Therefore,meaningless idle time is reduced.A customized file scheduler is also drafted to improve data layout dynamically at little cost.In order to fully verify the energy-saving effect of the new management strategy,this thesis conducts many experiments via simulator and real stoarge server respectively,using the real data of the AST3 telescope with detailed comparisons in terms of energy consumption and disk stability.Experimental results show that it reduces energy consumption by up to 56.6% and cuts down the switches of power state by up to66.8% compared with common energy-saving strategies which utilized a fixed idle time threshold.