Skip to Main Content
In this paper, we present a new software RAID technology called file RAID targeted at cluster environments where nodes share files through SAN (Storage Area Network). In it the file system manages separate disks and places user data over them redundantly by itself instead of by underlying device drivers while replicating the metadata on multiple nodes. This schema results in high performance by adopting the following techniques only applicable to the file system layer such as aggressive caching, reduced recovery logging and dynamic RAID type selection based on file size. It also eliminates conventional spare disks. We implemented the file RAID feature in our cluster file system HAMFS. The measurement result demonstrates a good scalability as well as good performance even for random writes for a RAID5 file.