Extending Lifetime Of Solid State Disk Arrays Based On Data Compression

NAND Flash-based Solid-State Drive(SSD)technologies have grown rapidly over the last decade.SSDs advance conventional magnetic Hard Disk Drives(HDDs)in size,performance,power consumption,low failure rate etc.,and are taking place of HDDs in production systems.However,the erase-before-write characteristic and limited erasure times significantly shorten the lifetime of SSDs.In particular,when deploying SSDs as an array and the workload is dominated by write requests,intensive writes and parity data updating wears SSDs quickly and greatly shorten SSD lifetime.A large number of technologies bring auxiliary Non-Volatile Memories(NVMs)and try to prolong the SSD endurance,but NVMs are expensive and even introduces additional reliability issues.Motivated by the fact that the abundance of data temporal redundancy is not uncommon between old and new data,compressions between them will considerably reduce the amount of write data and significantly mitigate the flash memory wear-out.a method of extending the solid state disk array based on compression is proposed.When writing dirty data,this method first does XOR operations between this new data and its old version,and then compresses the XOR result,finally writes this compression to a cache in sequence while doesn’t immediately update the corresponding parity data.By compressing data and lazy updating for parity data,this method significantly reduces the write stress on parity SSDs in the array,and therefore prolongs SSD lifetime.To note that this method uses a part of the SSD RAID,to ensure the reliability of compressed data,as the cache that stores compressions.This paper presents a self-developed array controller based prototype to demonstrate effectiveness and efficiency of this method.The results of the widely used benchmarking and real-world workloads show that,the SSD write stress is reduced by up to 80% when compared with traditional SSD arrays,and the performance is improved by up to 2.5x.