To accomplish this Gridstore abstracts the storage processing and storage capacity by moving the actual storage processing to the connecting clients, be it a server or a user workstation. What this means is that when a client needs to write data, the client itself actually parses the data and writes it across the available storage nodes. In short processing scales every time you add a user or server to the network. As a result the nodes that hold the capacity can be very modest in their processing power which should down the cost of storage significantly.

This model also addresses another challenge that the mid-sized data center faces: scale. As the business grows more and more file servers need to be added. Each of these need to be managed independently. With Gridstore you can simply add another node and the environment adapts to use the new node and its capacity. Also as stated earlier the processing performance of the NAS grows each time you add a user or server.

In order for the user system or server accessing the storage to be able to do this parsing of data a driver needs to be installed in the Windows kernel. Gridstore reports this to be relatively lightweight, 300k of memory when in use, and may consume as little as 10% of CPU on a lower end system making a large write to the grid.

Data protection is handled on a per volume basis. You can select a RAID 5 style of protection and be protected from a single node failure or via an administrator selection set the protection level higher to sustain multiple node failures. Defining per volume provides a quality of service like capability depending on the type of data being written.

Integration should be straight forward for the target market. To the user and administrator Gridstore, when set up, looks like a typical Windows file share. The grid can be managed by MMC plugins so administration and monitoring should plug right into the existing IT workflow. Today the environment is Windows only but Gridstore is working on drivers for other platforms.

Storage Swiss Take

From our perspective we’ve heard the “focused on the mid-sized data-center” claim before but Gridstore seems a little different. There is clearly processing power going unused on the connecting clients so moving some of the processing load to them makes sense to us. It certainly leads to a model that should drive down the cost of clustered storage while still providing the scale in capacity and performance that appeals about clustered storage. The combination we think achieves the Gridstore goal of bringing clustered NAS storage to the mid-sized data center.

George Crump, Senior Analyst

ExecEvent Briefing Note

Gridstore is a client of Storage Switzerland

First up was Gridstore, a grid based NAS solution whose goal is to bring scale out NAS functionality to the medium sized data center. The company was founded in 2007 and delivered its first beta in January of this year. They are attempting to solve a typical problem that the medium sized data center faces. They need more storage capabilities than a stand alone NAS can provide but can’t cost justify the expense of a more enterprise scale out NAS, which writes data across multiple storage nodes. To do this an enterprise scale out NAS needs to have sufficient storage processing capability to receive the inbound data, parse it and write it across all the nodes in the cluster. This process increases the cost and may put it outside of the reach of the mid-sized data center.