NIFFS: A New Index Flash File System for Network Applications

Abstract

Nowadays, big data has been one of the hottest research frontiers. It puts forward higher requirements to network storage systems. Flash memory has become a popular alternative to replace disks in the network storage systems because of its high performance. However, it has characteristics of erase-before-write and limited cycles of erase operations, which introduce new challenges into the design of file system. Conventional file systems do not consider flash memory characteristics, and bring extra writes to flash memory. We propose a new flash file system named NIFFS, to reduce the writes to flash memory and promote the performance on index. To reduce the writes, NIFFS redesigns data layout, inode structure and data organization. NIFFS promotes index performance by hybrid extendable hash-based directory structure (HyEx Hash) and adaptive file data index. In addition, NIFFS also provides metadata log to avoid frequent updating in metadata. Meanwhile, garbage collection with different granularities is used to reclaims obsoletes and a fast recovery which guarantees consistency by shrinking scanning space. The experimental results show that NIFFS reduces the index cost by 25% to 55%, and outperforms the state-of-the-art file systems both in throughput, read/write bandwidth and endurance.