Creating highly available data stores remains a significant problem. It is desirable that data should be highly available but - other than in the case of static, read only data - asynchronous changes make this very difficult to achieve. The advent of fibre channel and shared storage often referred to as a SAN15 has led to the development of The Global File System that effectively eliminates much of the need for file servers and single point of failure data storage.
The Global File System facilitates access to shared fibre channel disks without the need for a master node for mediation of resources. This is achieved by storing meta-data on the disks using device locks or dlocks. As there is no master node, there is no host acting as a single point of failure or bottleneck. While the disks themselves still represent single points of failure this can be eliminated by using a fibre channel RAID16 device. The failure of a fibre channel switch should, at worst, prevent access to part of the fabric.
GFS is currently functional and a network with in excess of 1Tb of storage was demonstrated at Linux World[2], New York in February 2000. Unfortunately journaling has not been implemented; while an array of n machines with m disks will work fine across the switched fabric, if a node fails then there is a risk that the file system will become corrupted. Work is currently in progress to resolve this problem and it is hopped that this will be completed before the end of the year.
GFS is developed by The GFS Group at the University of Minnesota, led by Dr. Matthew O'Keefe. The code is released under the GNU General Public Licence. More information can be found on the project web site17.