Akihiko Nishitani and Tomohiko Ogishi
Storage, distributed file system, mass storage
As basic technology for cloud services, distributed file systems are attractive due to the scale-out nature. Several sophisticated distributed file systems, such as Gfarm [1] and Hadoop [2] are actually in practical use, including commercial purposes by several organizations. However, due to the single point of failure (SPOF) at the center server, these systems are mainly applied to systems whose troubles do not impact on end users, which are backup systems and log collection systems. In this paper, we propose a non-stop distributed file system in order to apply them to not only such background applications but also many foreground applications. Our system creates redundancy by replicating I/O system calls by user operations from the clients and performing I/O system calls at multiple sets of distributed file systems in parallel. The experimental evaluations demonstrate that our system works efficiently with little overhead on consistency checks and error recovery.
Important Links:
Go Back
