Tuesday, May 17, 2011

Why the need for Scale in the SMB NAS?

The NAS is not only the first server and storage system that the SMB is likely to buy, it is also the first system that the SMB is likely to need to upgrade. Other servers that the SMB buys and uses may be appropriately sized to handle years of growth, but demand on the NAS is different. First, as the business matures more and more data is created, and the bulk of that data growth comes from office productivity applications like word processing, spreadsheets and presentations as well as rich media like podcasts and video. All of these consume storage space rapidly. As the SMB adds employees the rate of that growth also increases, since each new employee means more data generators and collaboration. More collaboration means that more copies of files are created prior to them being edited or modified, again storage capacity is further consumed as a result. The need to expand the storage capacity of the original NAS purchase to levels well beyond what was ever projected becomes critical to sustaining the business.

The use of office applications is key for the SMB. Presentations are used to create proposals, spreadsheets are used to track projects and word processing documents are used for outbound communications. Collaboration between employees speeds the development of this collateral and the sharing of these files becomes essential. If the NAS device goes down data may be lost and almost as importantly, the flow of information is interrupted. Making sure the NAS device stays up and running while it remains cost effective becomes as an important of a priority for the SMB as capacity expansion.

The SMB NAS Shortcoming

The problem is that most SMB NAS solutions are focused primarily on meeting a price point. They either come with a fixed amount of capacity or have limited internal expansion. Very few have the ability to expand beyond just internally mounted drives. This becomes a problem for the growing business as they very quickly need capacity beyond what the original NAS provided, but don’t need a higher-end, enterprise NAS. Remember, for the SMB, the primary goal is to store files that are needed for collaboration, not to run business applications or virtual machines.

In addition, each SMB NAS offering typically has many single points of failure (SPF). Most have one power supply, one RAID storage controller, one NAS server and even a single network interface. If any of these SFPs do indeed fail the entire system is down and the collaboration process comes to an abrupt halt. In this regard, reliability may be more important to the SMB than it is to the enterprise, where there is an IT staff available to monitor and fix problems as they occur. In the SMB there are typically no formal IT personnel or very few, and they are busy performing many functions. Having to restore a failed NAS from backup means pulling them off of another project, potentially for hours if not days.

At some point the SMB does need to expand beyond the limits of the original NAS. The only workaround available to them is to add a second NAS and to continue adding NAS devices each time another capacity roadblock is reached. The problem with this strategy is that it becomes a management burden and adds significant cost. Knowing where to store files, how to find files and how to protect files becomes increasingly complicated as each NAS is added to the environment, and complexity is the bane of organizations without dedicated IT staffs, like most SMBs. In addition performance of the environment is still limited to the capabilities of each individual NAS. Finally each NAS is its own stand-alone island - and single point of failure. If that individual NAS fails the data on that device is no longer accessible to users until it’s recovered or repaired, even though the other NAS systems may still be operational. In short capacity, performance and availability cannot be aggregated across the multiple NAS systems.

Performance scaling is also an issue, even in the SMB. To keep costs down the traditional SMB NAS solution uses low end processors. While at first this seems like a good practice, limited processing begins to show its weakness as the SMB grows in size. For example it takes processing power to move data to and from the 1GbE network that connects users to the NAS. Most of these processors can only fill about 20% of that bandwidth. Again, to keep cost down most NAS systems count on software to deliver functions like RAID, snapshots and replication. While doing these via software makes economic sense (software usually costs less to develop than hardware), each of these functions also consumes processing power. If they are all active overall performance may suffer dramatically. A good example is when a drive fails and the RAID software needs to rebuild the data that was on the failed drive. Because of the lack of processing performance the rebuild can take days.

Some single NAS systems, thanks mostly to larger disk drives, now have a very high fixed capacity, and the need for additional NAS units to meet growth demands may be minimized. The single, high capacity NAS actually amplifies the other problem areas of reliability and performance. If all the users are accessing the same NAS the performance capability of that device needs to be greatly enhanced. A single or dual 1GbE connection may not be enough to deliver data fast enough to the users, even though the system’s limited processing power may not support it anyway. Finally, if that single NAS fails, all the data in the environment and all the collaboration it supports fails with it. In short something is needed for the SMB that addresses the requirement to scale all three vectors: capacity, performance and reliability.

Addressing Scalability with Scale-Out NAS

The problem of scaling capacity and performance while maintaining reliability is a classic one for NAS systems of all sizes. But it’s exacerbated in the SMB because of the need to meet an acceptable price. In larger businesses scale-out NAS has stepped in to meet the data growth demands of the enterprise by providing linear capacity and performance expansion. These systems are typically 1U servers, called “nodes”, that are connected in a grid-like fashion and appear to the user and administrator as a single entity, thanks to a clustered operating system.

Each node has a set amount of capacity and performance (network and CPU), plus internal redundancy. When nodes are added to the cluster their capabilities are aggregated with the other nodes to provide the linear scaling that the business demands.

The challenge that scale-out NAS has had in appealing to the SMB market is that it’s expensive and somewhat of an overkill for that space. For example, most enterprise, scale-out storage vendors publicly state that 50TB is their minimum configuration and 75TB is often more appropriate. Also, many enterprises want to do more than just share files with their NAS infrastructures, they want to host virtualization environments and even databases on them. This means that the enterprise scale-out NAS has to use the latest, most powerful processors, provide a lot of network connectivity and use high performance drives. The use of these components requires more expensive server hardware and consumes more power and cooling as well.

Taking Scale-Out NAS to the SMB

The SMB needs many the capabilities of scale-out NAS, at least in terms of scale, performance and reliability, but at a cost it can afford and without the complexity of having to deal with things such as clustered operating systems. Companies like Gridstore are stepping in to fill this void with their NASg scale-out NAS for SMBs.

In a recent article Storage Switzerland detailed how scale-out NAS could be designed for the mid-range market, and those same techniques apply to the SMB market as well. Most of this centers on the use of a lower power, lower performing CPU based on the Intel ATOM processor. While maybe not suitable to host a virtualization environment or database it’s ideal for file sharing, especially when processing power is aggregated with other ATOM based nodes in a scale-out system. This design allows for very cost effective nodes while at the same time creating a denser, more cost effective packaging. As a result products like the NASg can be offered at lower price points and maintain a lower cost of operation, thanks to more efficient power and cooling, while offering scaling capacity, performance and reliability that scale-out systems are known for.

In addition, the NASg takes performance scaling a step further by off-loading some of the node management overhead to the connecting clients. The average client that connects to a NAS, thanks to today’s processing capabilities, typically has CPU cycles to spare. By installing a virtual controller on clients and servers each node in the cluster can handle some of the inter-node communications that burden larger systems. This is ideal in the SMB use case. Installing an agent to access storage across 5,000 laptops and servers and making sure that each new devices has that agent could be prohibitive in a large enterprise. On the other hand, installing an agent on a few dozen laptops and desktops and making sure that the occasional new system has the agent loaded is very manageable for most SMBs. NASg automates the agent install task by leveraging the SMB’s Active Directory and Domain Controller to make this a single push distribution.

Scale-out NAS may be more valuable to the growing SMB than it is to the enterprise. SMBs have the need to grow while maintaining acceptable performance and, as stated earlier, may have an even greater need for reliability, all of which scale-out NAS addresses. Solutions like Gridstore’s address the scale-out NAS’s cost issues and finally bring all the capabilities of scale-out NAS within reach of the SMB.

George Crump, Senior Analyst