Research On Cache Management In SSD-based Storage Systems

Limited by their own mechanical properties,HDDs have become the main performance bottleneck of storage systems.Different from HDDs,SSDs are flash-based devices with higher read and write performance,lower energy consumption and better shock resistance than HDDs.In recent years,SSDs have been applied to many environments,such as personal computers,mobile devices,VM-based storage systems and high-performance enterprise storage systems.However,cache management algorithms in traditional storage systems are customized for HDDs.Applying SSDs directly to traditional storage systems will cause some problems.Therefore,it is necessary to optimize the cache management algorithms of storage systems based on the characteristics of SSDs.In this paper,we propose two new cache management algorithms by deeply exploiting the characteristics of SSDs.Firstly,we propose a Garbage Collection aware cache Replacement algorithm,called GCaR,to improve the performance of flash-based SSDs according to the Garbage Collection(GC)characteristics of SSDs.GC is an important performance concern for SSDs and brings a negative impact on the Quality of Service(QoS)of SSD-based storage systems.In order to address the performance issue caused by GC operations,GCaR gives higher priority to caching the data blocks belonging to the flash chips in the GC state to lessen the contention between user I/O requests and GC-induced I/O requests thus reducing the average response time of user I/O requests.The simulation results show that GCaR significantly alleviates the performance degradation caused by GC operations,and improves the performance and QoS of SSD-based storage systems.Secondly,we propose a Multi-channel Parallelism aware cache Write Back algorithm,called MPWB,by exploiting the multi-channel parallelism within SSD.Usually,the dirty data in the cache is written back to SSDs when being eliminated.New data cannot be written into the cache until the eliminated dirty data has been written back to SSD that would increase the average response time of user I/O requests.When flushing the dirty data from cache to backend SSD,MPWB writes back multiple cached dirty data belonging to different channels in parallel,which only takes the same write-back overhead as a traditional write back operation.The simulation results show that MPWB effectively reduces the cache write-back overhead and improves the performance of SSD-based storage systems.These two key technologies of cache management fully exploit the garbage collection characteristics and internal parallelism of SSDs respectively to optimize the cache management algorithms,thus improving the performance and QoS of SSD-based storage systems.