Network Computing is part of the Informa Tech Division of Informa PLC

This site is operated by a business or businesses owned by Informa PLC and all copyright resides with them. Informa PLC's registered office is 5 Howick Place, London SW1P 1WG. Registered in England and Wales. Number 8860726.

Scale-Out Storage For Hyper-V

This is the second post examining products I saw at Storage Field Day 4. held earlier this month. In my first post, I focused on Overland Storage's SnapScale. While SnapScale is as conventional as a scale-out storage system can be, Gridstore’s eponymous product not only uses an innovative host-side virtual controller architecture, but is also the first storage system specifically targeting Microsoft Hyper-V infrastructures.

Most scale-out storage systems use a standard protocol such as SMB or iSCSI for access. The problem is that these standard protocols weren’t designed for scale-out, so storage system designers have to figure out how to deal with requests that come into one node when the data lives on another and also how to replicate data between the nodes for data protection.

Rather than use a standard storage protocol, Gridstore places much of the work of managing the scale-out cluster into the client as a virtual controller. The virtual controller installs into Windows as a Storport miniport driver, making the Gridstore nodes available to Windows servers as block storage just like iSCSI, Fibre Channel or DAS.

The virtual controller chunks up the data and stripes it across the storage nodes using a Reed-Solomon- like erasure coding so users can set their protection level from 1 to n drive and/or node failures with lower overhead than n-way replication. All I/O is direct from client to storage nodes, so no backend network is required.

Gridstore’s data nodes are based on a Dell 1u server with four 3TB SATA drives. The C-Class nodes have six 1Gbps Ethernet ports while the IOPS oriented H-Class adds a PCIe SSD and replaces a pair of the 1Gbps ports with the 10Gbps variety. Using Dell servers allows Gridstore to leverage Dell’s service and support infrastructure, which of course has spare parts pre-staged around the globe.

Like the systems they support, the storage nodes also run Windows with the H-Class nodes using SanDisk’s FlashSoft to manage the PCIe SSD as a write-back cache. Since data is simultaneously written to multiple nodes, there’s no need to specifically protect the dirty data in any given node's cache via replication or other methods. Gridstore claims that a small cluster of three H-Class nodes can deliver 38,000 8K random IOPS. Since larger clusters don’t create more overhead with internode communications, the system should scale pretty close to linearly.

[Read about the improved storage functionality in the latest Windows Server release in "Windows Server 2012 R2 Beefs Up Storage Spaces."]

I first saw Gridstore about two years ago, when now CTO Kelly Murphy was pitching the technology as a scale out NAS/file services solution for the small business market. While I’ve always liked the idea of scale-out, Microsoft’s SMB protocol worked just fine for PC to file server access, and I couldn’t see enough advantage to Gridstore’s approach to make it worth installing a virtual controller driver in hundreds of PCs.

It’s in the Hyper-V environment where the host-based virtual controller starts to make a lot of sense. Since the controller runs in the host, it can detect which IOs are targeting each virtual disk (.VHD or .VHDX file) and the Gridstore system can provide all of its data services from snapshots and replication to thin provisioning on a virtual machine by virtual machine basis rather than volume by volume basis.

More significantly, it allows Gridstore to provide end-to-end quality of service on a per VM basis. As I’ve written before, I firmly believe that as we consolidate servers and storage into fewer pools, storage QoS will become the defining feature for next-generation storage systems.

I find it fascinating how different vendors can create unique products from the same basic parts bin of commodity servers and disk drives. If, as I expect, Hyper-V continues to grow, Gridstore should be able to take advantage of its position as the first Hyper-V optimized storage system.

Disclaimer: I went to Storage Field Day, travel paid, and all I got was a lousy T-shirt (OK, several T-shirts and some other minor swag). Gridstore is not a client of DeepStorage, LLC though I do have some ideas for a project with it that involves pyrotechnics.