October 22, 2014

Scaling Storage At The Client

Wouldn’t it be great if storage protocols allowed for scalability?

All three of the array-scaling approaches I discussed last week preserve traditional client protocols for connectivity, an admirable goal to be sure. But the inflexibility of these protocols holds the storage industry back, often requiring extreme efforts to achieve simple goals like continuous availability. Attempts to evolve protocols to be more flexible have met with a great deal of resistance, but this hasn’t stopped companies from trying new concepts.

Rather than trying to force existing protocols to scale, many companies are turning to the client. What if we used a special driver or even an entirely different protocol? One common approach to scale the client is to use a special storage driver that provides connectivity to multiple storage nodes.

This is part of a series on “Scale-Out” Storage Field Day 4

  • A multi-path I/O (MPIO) driver installed in the operating system can provide high availability and a bit of flexibility to boot. I’ve relied on MPIO from Veritas (DMP), EMC (PowerPath), and Microsoft for decades, and these continue to be used in today’s operating systems and hypervisors (see VMware PSP, for example). But today’s MPIO drivers are really intended to provide availability, not the kind of scalable storage we’re thinking of in this article.
  • Alternative storage protocols like ATA over Ethernet (AoE) can provide high performance, scalability, and reliability through tight integration between clients and storage nodes. But not everyone is comfortable with such a radical change to client storage connectivity. Consider the fate of pNFS, an industry-standard scalable NAS protocol that has seen almost no client adoption despite being available for most of a decade.

Gridstore

At Storage Field Day 4, the delegates learned about Gridstore, which uses a “virtual controller” installed in each client (typically a Hyper-V host) to distribute data across many independent nodes. This reminded some of the Coraid solution, presented at Storage Field Day 1, which uses a custom host storage controller to connect to storage nodes using the AoE protocol.

Because Gridstore is currently Windows-only, the company has wisely decided to focus the product on Hyper-V storage. It is rather well-suited to this application, offering scalability of capacity and performance with 1U nodes. If anything, the Gridstore appliances offer excessive CPU capacity, suggesting a converged offering (like those recently announced by Violin, Scale Computing, and EMC) might come next.

Oxygen Cloud

Another “change the client” option is application-integrated cloud storage. Oxygen Cloud used Storage Field Day 4 to introduce a new end user-focused enterprise file sharing product called “odrive”. Oxygen places a gateway in front of an enterprise file server or NAS array (IBM Storwize, in their demonstration) which offers storage directly to client computers through a proprietary client.

odrive functions much like Dropbox but uses in-house storage along with a cloud component to facilitate connectivity and synchronization. Since the odrive client can transparently connect to many gateways, Oxygen offers simple scalability and geographic independence.

Cleversafe

Cleversafe first presented at Storage Field Day 3, outlining their scale-out object store architecture. This time around, Cleversafe focused on use cases for this technology, including a discussion of their combined gateway solution with Avere. But Cleversafe is a hyper-scalable storage array on its own, with a REST interface allowing simultaneous access to a true storage cloud. Cleversafe even has their own Accessor “data routers” as well as a client application allowing direct access to the cloud.

At Storage Field Day 4, Cleversafe spent a lot of time talking about customer use cases in a variety of scenarios, including using gateways from Panzura and Avere as well as media distribution with ctera. They also demonstrated scaling and data distribution, as seen in the video below.

Stephen’s Stance

Scaling storage is a serious challenge for the industry, but there is a great deal of thought, effort, and creativity going into it right now. Companies like Gridstore, Oxygen Cloud, and Cleversafe have come up with effective client-side solutions to enable scale-out storage to sing. If you’ve got an appropriate application, client, or gateway, scale-out is a real possibility!