AAS: Address Aware Scheduler for Enhancing the Performance of NVMe SSDs in IoT Applications

Abstract

The use of NVMe SSDs have been growing in many applications as well in IoT due to their high-performance features as compared with the traditional SATA SSDs. Although, NVMe SSDs provide higher performance due to their multi-Queues feature, but this feature may not be employed efficiently due to the wrong configuration as well as weak scheduling. The scheduler plays a key role in handling the requests to separate among multi queues. Traditional schedulers usually fulfill the queues of SSDs with aim of providing parallel processing. While, in some conditions, especially in IoT applications, many useless repeated writes have been issued to the memory which may degrade the performance. In this paper, we first investigate the main parameters which affect the performance of NVMe SSDs; then, we proposed an address aware scheduler caller AAS, which improves the performance of SSDs by mitigating the useless writes. The results of evaluation on MQSim shows that the enhanced performance of NVMe SSDs will be achieved in specific configuration when running simulator. By considering those configurations, we evaluate the performance of SSDs when using proposed AAS scheduler, the results show that the proposed method enhances the performance significantly (up to 111%) especially in the cases with high percent of repeated data.