MatBall: Matrix-Based File Allocation Method for Hybrid File System
Jaechun No, Sung‐Soon Park · 2013
One way of improving I/O performance is to provide a hybrid file system where the entire address space is provided by combining a small portion of SSD (Solid-State Device) partition with a much larger HDD storage capacity. Since the address space to be offered by SSD partition is very small, utilizing SSD storage resources is the critical aspect in improving I/O performance of the hybrid file system. We present a matrix-based file allocation scheme, called Mat Ball (Matrix-Based file allocation), whose objective is to increase space utilization of SSD partition in the hybrid file system by reducing fragmentation overhead. In Mat Ball, to consume the remaining spaces as much as possible posterior to file allocations, I/O units (extents) of the hybrid file system are recursively partitioned into segments in the subsequent level and further file allocations are performed in units of the partitioned segments. Since Mat Ball defines easy-to-compute segment sizes and block positions in I/O units, allocating more files in the remaining spaces can be performed with a little overhead. The performance measurement with IOzone shows that the hybrid file system using Mat Ball enables to produce higher bandwidth over ext2 installed on HDD and SSD.