Centralized processing, also known as the mainframe, has had a rocky existence of late. First it was in, being the only game in town. Then it was out as distributed architectures using low-cost blade servers took hold. Now it's coming back in as enterprises ramp up for the cloud.
Could it be that the central storage array is following the same path?
Popping up on some "2010 Trends to Watch" lists already is the rise of clustered storage, also known as grid or scale-out storage. The idea here is largely the same as in distributed server architectures: Replace the central behemoth that is the storage array with low-cost storage nodes. In that way, you can add or remove capacity more quickly and easily or even scale one component of storage, say processing, while leaving capacity and bandwidth alone, says InformationWeek.
One of the chief requirements of clustered storage is, naturally, a robust network infrastructure. That's why some firms are loading their platforms with high-end technology that otherwise would be priced out of an integrated system. The Register says IBM, for example, is loading its XIV platform with InfiniBand, the better to shuttle data across the up to 15 79 TB nodes that the system offers, soon to be upgraded with 2 TB drives for an effective capacity of 160 TB per node.
At the same time, smaller firms are looking at clustered storage as a wedge to target small and medium-sized organizations. An Irish start-up, Gridstore, is building its NASg node around the Intel Atom processor, providing 1 or 2 GB per node for about $400 each. That company is hoping to draw the interest of managed service providers, arguing that its approach outclasses the smaller NAS offerings from Iomega or NetGear.
And since the lowest-cost way to set up a storage grid architecture is through commodity hardware in the node, a number of firms are hitting the market with open source management software that can easily integrate into mixed environments. A company called Gluster recently added an open source file system, GlusterFS, to its storage OS and management system. The Gluster Storage Platform's claim to fame is that it eschews the use of a central metadata server, arguing that it can avoid the performance bottleneck caused by centralization while overcoming management issues by spreading network intelligence out to the node.
Just as the predictions over the mainframe's imminent demise have proven to be wrong, it is certainly premature to argue that clustered storage is about to wipe the traditional storage array off the map. But for those expecting to see rapidly changing storage requirements over the next few years, it's at least nice to know that there is more than one way to add or remove capacity in a timely fashion.