Open source file system vendor Gluster released its Storage Platform, integrating a file system, an operating system and a management UI to create a cluster that can store petabytes of information and provide failover for virtual machines.
The platform is designed to run on commodity hardware just like GlusterFS file system, which is at the Storage Platform's heart. The operating system is a stripped down Linux kernel and the management interface was built from the ground up by Gluster, which was named last month by Network World as an open source company to watch.
"Previously we had a command line interface and a minimal GUI for management so we built a new tool," says Jack O'Brien, Gluster's senior vice president of marketing.
GlusterFS is a clustered file system for storing unstructured data. It features a global namespace, but does not have a centralised metadata server. Instead, GlusterFS uses an index to look up files, employing a hashing algorithm to find a unique identifier for each file. Each server in the cluster knows where every file is stored.
The file system, however, is not just for storing data files. Users also can store multiple copies of virtual machine images and use the unique features of Gluster Storage Platform to create failover capabilities for the VM environments.
"What we changed in the file system was in the area of always-on virtual machines," O'Brien says. "We are great at managing files, and VM images are just files."
The Storage Platform is designed to withstand a hardware failure and allow users to recover VMs without disrupting the services running on the Storage Platform.
To do that, Gluster does checksum-based healing. "We compare the live file to the copy and that allows you to figure out the error within the file as opposed to working at the file level," O'Brien says. "So you don't have to rebuild the entire file. VM images can be very large so stopping to rebuild it can be very disruptive."
To build to scale, Gluster takes disk memory resources and aggregates them into a single pool under one global namespace. That centralisation facilitates a single management model, company officials say. Gluster then presents only one IP address that every virtual machine in the cluster talk to. The system automatically distributes the I/O and data across the cluster.The Storage Platform also features web-based installation and management and a complete platform image that can fit on a USB thumb drive.