STAR: A Zone Translation Scheme to Improve the Performance of Host-Aware SMR

The exponential growth of big data requires more and more storage space. Host aware SMR (HA-SMR) drives are the most promising among the three types of SMR drives which can effectively improve the areal density of hard disk drives. However, when handling non-sequential writes under write-intensive workloads, a host-aware SMR drive often suffers from performance degradation. In this paper, we propose a hybrid storage system called STAR, which consists of an SSD cache and an HA-SMR drive. In STAR, we propose a new data layout management for the HA-SMR drive, which divides the whole space of the HA-SMR drive into data zones and a log section. The log section is used to buffer data blocks for the data zones. We design a zone translation layer (ZTL) at the host side for the HA-SMR drive to translate non-sequential writes to sequential ones, and we call it ZTL HA-SMR drive. In addition, we also use an SSD cache to absorb non-sequential writes for the ZTL HA-SMR drive and design a cache replacement algorithm called LCF, which can reduce data migration in the ZTL HA-SMR drive. Our experimental results show that STAR can effectively reduce non-sequential writes to the HA-SMR drive and improve the performance of the HA-SMR drive as well.