Introduction to Linux Cluster Filesystems

Traditional local file systems support a persistent name space. A local file system views devices as being locally attached, the devices are not shared, and hence there is no need in the file system design to enforce devicesharing semantics. Instead, the focus is on aggressively caching and aggregating file system operations to improve performance by economizing on the number of actual disk accesses required for each file system operation. Newer networking technologies allow multiple machines (nodes) to share storage devices. IBM’ General Parallel File System (GPFS) or Red Hat’ Global File System are representing distributed file system technologies that are taking a shared, network-attached storage approach. These file systems are built on the premise that a shared disk file system has to exist within the context of a cluster infrastructure, and has to provide proper error handling and recovery, as well as the best performance possible (performance, availability, and scalability features are key requirements). In a SAN attached environment, SAN clients may only manage local file system requests and act as file managers for there own I/O operations. Hence, the storage devices (the IO subsystem) serve the data directly to the clients. A cluster file system design provides transparent parallel access to storage devices while maintaining standard UNIX file system semantics. User applications only view a single logical device via the standard open(), close(), read(), write() and fcntl() primitives. This transparency is paramount in regards to ease of use, as well as to the portability of these file system technologies. To reiterate, a cluster file system based design differs from traditional, local file systems, by emphasizing sharing, connectivity, as well as (client side) caching. Unlike local file systems such as IBM’s J2 or SGI’ XFS, cluster file systems distribute file system resources (including the metadata) across the entire storage subsystem, which allows simultaneous access from multiple machines.