GlusterFS is a distributed file system presents the memory elements of a plurality of servers as a single file system. The different server, even cluster nodes (English node, node ') called, form a client -server architecture over TCP / IP. A special feature of NAS systems can be integrated via Infiniband directly into the cluster, also a redundant connection of storage devices over TCP / IP, Infiniband Verbs InfiniBand or SDP ( Socket Direct Protocol ) is possible. The data on all cluster nodes can be read and written at the same time, all changes will be implemented immediately to files on all servers. The file system is mounted on a FUSE kernel module and is supported by POSIX -compliant operating systems, including Linux, FreeBSD, OpenSolaris and Mac OS X. To start a GlusterFS server, a kernel module is needed. A server can be both client and server must be the same. A client for Windows systems is in the planning stage, it is being implemented by the developers only once the WinFUSE project is stable.
The development of GlusterFS began in mid-2005 by the GlusterOS Development Team Z Research Inc., a first release of the file system was released in July 2006. GlusterFS is licensed under the GPL version 3. The developers offer paid support. The end of 2011 was GlusterFS Red Hat U.S. dollar bought for 136 million.
GlusterFS has a modular design and supports multiple modes of operation:
GlusterFS uses a modular design, so that all described functions can be mapped with few components. The components are data partitions ( volumes ), transport groups ( transport groups) and translators ( translators ). The translators provide the opportunity to expand the file system functions, in particular whether it should communicate via Infiniband, TCP, or a mixture of both, which physical partitions or directories to be included and which RAID levels are used. To illustrate the modular concept, a ROT13 translator exists to " encrypt " the file system.
With GlusterFS can be a kind of network RAID create, from which multiple computers can access a shared file system simultaneously. Here it is not subject to limitations, such as the being able to use a maximum of two servers. GlusterFS is fault tolerant because can be stored in distributed GlusterFS user data, metadata and namespace. Through each other GlusterFS server increases the maximum data throughput of the file system, so that I / O bandwidth can be achieved by some GiB per second here.
For processors that Moore's law, but this is not the case in storage media and storage solutions considered, although there is also a need for larger and faster memories. Often is not the CPU performance of a server is the bottleneck, but increasingly to the slow data memory of the system. GlusterFS provides a remedy by the ability to scale to any size.