The rationale for deploying a clustered storage system is in many ways similar to that of deploying clustered servers: You get better scalability, both for capacity and performance, and more resilience than traditional solutions can provide.
Not surprisingly, the traditional adopters of clustered storage are companies with applications that push the limits of traditional file systems. That's a rather large and diversified group of applications that includes, for example, collections of scientific and medical data but also more mundane activities like managing large amounts of videos or music files.
Today, you probably won't find Joe in Accounting among the users of clustered file systems, because using such high-capacity, high-performance solutions for bookkeeping would typically be overkill. A conventional NAS system would probably be the best fit for those requirements.
It's no mystery that NAS solutions are becoming faster, but their limitations on capacity remain, which is why new developments in distributed file systems have further widened the performance gap between NAS systems and the distributed file systems offered with clustered storage.
The latest version of Isilon's clustered storage solution, OneFS 4.5 (stands for One File System), announced Oct. 9, continues that trend, bringing the maximum capacity to a resounding 1PB (petabyte) per volume and claiming an aggregate transfer rate of 10GBps.
You may remember Isilon from previous from previous articles. As a quick recap, OneFS is a distributed file system that runs on multiple nodes of proprietary hardware, connected via iSCSI links to their hosts and interconnected via iSCSI or InfiniBand.
Of course, "distributed" means that OneFS will automatically allocate file space across all nodes and all spindles belonging to the same cluster, while maintaining a single namespace and a single management view.
In plain language, you can manage the system from any node and maintain a consistent global view. Hooking up more nodes or more spindles will automatically increase performance and capacity.
In addition, the Isilon system offers features such as remote synching and multiple redundancy. For example, OneFS can survive the simultaneous failure of as many as four nodes, and provide exceptionally fast rebuilds of failed drives.
The additional capacity that the new version offers did not require re-coding: "We don’t have a capacity constraint burned in our software," explains Sam Grocott, director of product management, adding that OneFS has theoretically unlimited capacity but has been tested to as large as 1PB so far.
OneFS performance numbers will turn heads, but I am more impressed by a new feature, snapshots, which was until now conspicuously absent from Isilon's management apps.
Aptly named SnapshotIQ, this new feature allows customers to take an unlimited number of snapshots per system. I'll repeat, there is no limit to the number of snapshots you can take cluster-wide, but you can take "only" 1,024 for each directory.
Even 1,024 snapshots per folder (plus the reliability of the system) should keep exposure to possible data loss to a minimum. Moreover, space for snapshots is allocated system-wide, which should also simplify management.