An Efficient, QoS-Aware I/O Scheduler for Solid State Drive

Flash-based solid-state drive has been proved to be a competitive alternative to mechanical disk for its high performance and low power consumption. But SSD still suffers from relatively high price and low density, which calls for consolidating device resource to bring down the cost of deploying expensive dedicated flash-based storage systems for specific applications. Within such environment, applications may compete for storage service and interfere with each other. Therefore, storage service providers must ensure performance isolation. However, current research mainly focuses on performance guarantees on mechanical disk. In this paper, we present a 2-level scheduling framework where the higher level targets performance isolation through distinguishing the cost of read and write operations in SSD, and the lower level builds channel-based queues for SSD to exploit the inherent parallelism of SSD. Synthetic and real workloads are used in our evaluation to demonstrate the effectiveness of our 2-level scheduling framework, and the results show it outperforms other I/O schedulers at both aspects of performance isolation and I/O efficiency.