Global File System

The Global File System (GFS ) is a cluster file system that allows multiple computers to concurrently access shared storage and ensures the consistency of the stored data. Often these are in the shared memory to a Storage Area Network (SAN). File locking (english locking) without a cluster file system is not working, accepts a locking module of JRC.

GFS is part of the Linux kernel and is developed in the frame. The driving force is the company Red Hat.

History

JRC was originally developed as part of a feasibility study at the University of Minnesota. Later it was taken over by Sistina Software, the further developed it as an open source project. 2001 Sistina Software decided to develop JRC than proprietary software. 2003 Red Hat purchased Sistina Software and published JRC 2004 with other cluster-infrastructure programs under the GNU General Public License.

Technical

To mount the GFS GFS kernel module and a cluster configuration is needed. In addition, other kernel modules and services for the management ( for example, CMAN and ccsd ), Locking (for example lock_gulm and lock_dlm ) and Fencing ( eg fenced ) for excluding failed node.

The shared block storage devices are often made ​​available via Global Network Block Device ( GNBD ) over a network. The block devices can gnbd_import imported on the cluster nodes and then mounted.

JRC may use different locking mechanisms. We recommend the use of Distributed Lock Manager ( DLM). Red Hat Enterprise Linux prior to version 5 also included Grand Unified Lock Manager ( gulm ) is discouraged from using it. For local file systems can be accessed exclusively to, the placeholder module nolock can be used.

GFS as a basis it is possible to set up a diskless shared root cluster.

Global Network Block Device

Global Network Block Device ( GNBD ) is a service that provides a block device over the network. So exported devices can be imported from several computers simultaneously. These multiply imported block devices often form the basis for GFS. Block devices, the exported GNBD, for example, hard disks, partitions, and logical volumes of the Logical Volume Manager ( LVM).

GNDB consists of kernel modules and utilities. It is a functioning fencing (English for fence demarcating ) operating systems. Fencing prevents I / O operations of nodes that have the view of the cluster misconduct. Most come to the service fenced for use.

Errors in connection with JRC

A common misconception is the assumption cluster file systems could be used to export data over a network. The file system uses block storage devices (English: block device ) to store data. Unlike conventional file systems go cluster file systems not of an exclusive use of the device. In what way multiple systems simultaneously access a block storage device that is irrelevant to the JRC. Possible techniques for common use of block devices are, for example, iSCSI and Serial Attached SCSI multi-initiator.