Thursday, June 23, 2011

When data grows the SMB will usually add another NAS appliance, since most of these simple devices don’t scale. This growth ‘strategy’ can cause serious issues with reliability, performance and administration as each new box brings another single point of failure, another potential I/O bottleneck and another device to manage. There are scalable NAS solutions available, but they’re designed (and priced) for mid-to large-sized enterprises. This puts SMBs in a tough situation. They have some of the scaling and availability needs of much larger organizations, but the products designed to meet these needs are not designed for their environments.

They really need a NAS solution in between a low-end, fixed-capacity appliance and a higher-end enterprise solution; one that expands easily and offers higher availability - but is still affordable. In addition, this product needs to be simple to operate and implement, as SMBs don’t usually have a high level of IT expertise. Gridstore has a scale-out storage solution that’s designed specifically for the SMB space and addresses many of these requirements.

Gridstore’s NASg is a NAS system comprised of 1U nodes, each a complete storage module with disk capacity, processing power and network connectivity. The GS-1000 Storage Nodes come with 1TB or 2TB of disk storage, 2GB of DDR2 RAM, a dual- 1.8GHz core processor and GbE connectivity. As the name implies, a grid architecture allows users to scale the storage system by simply adding nodes. Similar in function to traditional scale-out storage the NASg offers the reliability, performance and expandability of systems built for the mid to larger end of the NAS market, but was designed from the ground up for the SMB.

Eric Slack, Senior Analyst

Nodes are made for a standard 19” rack, but are less than 10” deep and are quiet enough to sit on a desk. The NASg platform also offers thin provisioning, user quotas and performance monitoring with realtime alerts, but is controlled by a Microsoft Management Console (MMC) Snap-in, to make it easier for SMB customers to use.  The NASg provides the fault-tolerance and parallel I/O performance of a full scale-out architecture, but priced at $3795 for a 10TB stack.

Traditional NAS systems relied on a centralized processing architecture, which meant that growth was limited by the power of the controller. Scale-out systems addressed this  limitation by creating a modular architecture in which capacity and processing power were combined in each node. But this clustered design required that each node handle the usual storage functions, like RAID processing, file services and I/O, plus the additional overhead associated with physically distributed data sets. Scale-out NAS systems accommodate this with powerful and complex hardware, in systems that were priced way outside of the SMB market.

Given the recent history of microprocessor development, most computer CPUs are significantly underutilized. Part of Gridstore’s differentiation is the way it leverages the excess processing power that’s sitting in most desktops and end user computers. Instead of building a dedicated controller for the NAS system or simply moving that controller onto each storage node, Gridstore virtualizes the NAS controller with a piece of software loaded onto each end-user computer, making it a ‘Virtual Controller’. This architecture distributes much of the workload associated with managing data in a scale-out storage system, such as locating needed data blocks in the cluster, communicating with storage nodes and assembling files. By removing this overhead to the end user’s CPU each storage node can use lower power and less costly Intel ATOM processors. The CPU percentage consumed by the overhead for each Virtual Controller is in the single digits.

This architecture allows performance of the cluster to scale as Virtual Controllers (users) are added and provides a fault tolerance that’s more effective than RAID 5/6 and more efficient that RAID 1/0. Called RAIDg, a gridstore can be configured with as many redundant nodes as desired per volume, to sustain the required level of drive or other component failures. And, Gridstore’s distributed processing architecture prevents the performance impact associated with typical RAID 5/6 processing or rebuild cycles.

Below is a screenshot of the Gridstore MMC plug-in which can be snapped into any Microsoft management application and made available as another tool in the left hand panel. This dashboard shows the status of the entire gridstore. Operational statistics can be viewed at the storage pool, volume or storage node levels. Although currently healthy, warnings requiring immediate attention would show in yellow or red.

From the user’s perspective, Gridstore represents nothing new to learn, as capacity shows up as a new network drive letter, ready to go. The scale-out architecture produces dramatic performance levels from low-cost components. For example, the screen below shows a user writing data to the Gridstore, with a standard, low-end Dell desktop. Using a 5-node configuration, this client is seeing throughput over 60MB/s. Again, this is not shared throughput, but the performance one user is getting. In this architecture, the client is the bottleneck, so all users see similar or better performance.

Storage Swiss Take

File services are an early mission critical piece of IT gear for many SMBs, which often rely on fixed-capacity, NAS appliances to meet user demands. But stand-alone systems don’t scale adequately, resulting in the accumulation of multiple NAS storage units, each representing a single point of failure and an increased management burden. Until now the scale-out storage industry has left these small company users behind. Gridstore’s NASg provides a simple, scale-out file storage solution with surprising performance and an even more surprising price point. By leveraging the abundance of CPU resources that are sitting idle on most desktops, Gridstore has found an innovative way to remove much of the processing load from their storage nodes and keep the cost within the SMB budget.