A metadata cooperative caching architecture based on SSD and DRAM for file systems

Abstract

The metadata IO plays a critical role in achieving the high IO scalability and throughput to file systems. Due to the resource contention, the performance of the metadata IO is low. Adding the SSD into the storage system is a effective way to improve the performance of file systems, but the current methods mainly focus on the performance of the data server, rarely aim to the performance of the metadata IO. In this paper, we proposed a novel cooperative caching management algorithm based on DRAM and SSD named ACSH. By exploiting the temporal locality widely exhibited in most of the metadata workloads, ACSH can improve the performance of the metadata IO with reducing the write traffic to the SSD, and it includes a adaptive adjustment model, which can adjust the number of the cached metadata according to the locality strength of the metadata workload for improving the perforamcne and reducing the write traffic to the SSD cache layer further. ACSH has been evaluated based on the real-world workloads. Our experiments show that ACSH can reduce the latency by up to 1.5–3X in contrast with the original cache consisting of DRAM which has the same cost with ACSH. Compared with the recent study LARC, it can reduce the write traffic to the SSD by up to 23–30 %.