Design and evaluation of fault-tolerant shared file system for cluster systems

The paper describes the design and evaluation of a Fault Tolerant Shared File System (FTSFS) architecture for cluster systems with shared disks. The FTSFS architecture: guarantees no file system (FS) structure crashes on processor/program failure; can be applied to any existing nonshared FS without changing the structure of the FS; and does not degrade performance on the shared FS compared with a standard non shared FS. Using the FTSFS architecture, we implemented a fault tolerant shared FS on Fujitsu's SVR4 duplex system, and evaluated the system performance. The evaluation showed that the shared FS is competitive in performance with the standard SVR4-UFS (Unix File System).