Clustered network storage: part two; An evolution in storage.

The first part of clustered network storage discussed the general principles of this compelling architecture. Clustered network storage systems are an evolution of the two-way active-active architectures found with traditional midrange storage systems. Clustered network designs extend the number of intelligent controllers beyond just two controllers while appearing to the applications, users and system administrators as a single logical system. This article will analyze at a high-level what some of the potential capabilities customers can take advantage of with clustered network storage systems. In this article I refer to a controller as the intelligent head of the storage system that has processors, memory and most, if not all of the software that runs the entire storage system. A two-way active-active storage system as two controllers, a three-way has three controllers, etc.

[FIGURE 1 OMITTED]

"Infinite" Scalability

As discussed in part one, clustered network storage systems can keep scaling beyond traditional systems. Customers can start out with systems that have one or two controllers and keep adding controllers as needed. As you add more servers, users, or the number of transactions grows, clustered network storage systems allow you to just add another controller to the cluster that provides more processing power, cache memory and host ports to meet the demands.

With traditional storage systems when customers run out of hardware resources they have to buy a completely new storage system with the same limitations as the previous system. Customers begin to build more and more islands of individual storage systems adding to the complexity of storage management and additional capital expenditures. Clustered network storage is a single system so management stays fundamentally the same regardless of how many controllers you add. Additionally, capital expenditure should be less when adding to a cluster because customers are not buying completely new systems. Rather they are adding onto their existing system with less hardware costs and no additional software licenses.

ESG Lab tested a number of these solutions and performance grew as controllers were added in a near linear fashion. This proves that these systems have extremely efficient inter-nodal communication and can scale to massive amounts of performance. ESG Lab tested systems that grew to over 20, 25 and even 32 nodes and performed over one million, one and a half million to nearly two million inputs/outputs per second (IOPS). These results are staggering considering that most traditional high-end storage systems cannot claim these kinds of performance numbers.

Stretching Clusters

An extremely valuable capability that clustered network storage systems bring to customers is the ability to "stretch" the clustered controllers within campus distances. Many clustered network storage systems have the ability to physically separate controllers from one another. For example, customers can put two controllers in one data center and two other controllers in another data center across their campus. Why would customers want to do this? It results in having a simple and essentially free disaster recovery solution. If one of the data centers becomes flooded or has a fire, the two controllers at the other unharmed data center are still operational and have all of the data. If the customer also clustered their application servers between these two data centers, the users will experience no downtime. There may be an impact on application performance since half of the cluster would be unavailable.

The beauty of stretching clusters is that it comes for free, customers don't have to configure anything, there should be no data loss, and if a disaster occurs there is no downtime. Stretching clusters is not good for disaster recovery over long distances beyond FC or Ethernet limitations.

Extremely High Availability 

There is a concept referred to as Redundant Array of Independent Nodes or RAIN. Clustered network storage systems that support RAIN or RAIN-like functionality can survive multiple controller failures. For example, if one controller failed in a system that had four controllers in a single cluster the other three would continue on as if nothing happened. And some clustered network storage systems would be able to survive a second and third controller failure. These solutions combine RAIN and RAID offering extremely high availability to customers.

Single Level of Management

This is a powerful concept that you will see ESG writing more about in the coming year. A single level of management (SLM) combines all of the above to create an easy to use and scalable storage (network) system? that eliminates many of the mundane tasks that system administrators are faced with today. These tasks include capacity management, performance analysis and tuning and storage provisioning. These tasks are essentially eliminated with clustered network storage systems.

Alexander Graham Bell said that the great advantage that the telephone has over any other electrical device is that it "requires no skill to operate the instrument". A single level of management makes it essentially as easy to manage a storage system with 1 petabyte (1,000 terabytes) of capacity as it is to manage a storage system with 1 terabyte of capacity. The old rules in storage networking are changing, and as complexity is eliminated the market will grow even larger.

There are different approaches and nuances to the way clustered network storage is implemented depending on the product. Part three of this article will discuss the different approaches in a bit more detail covering SAN, NAS and Object-based solutions.

Part one of this article appeared in the January 2005 issue of Computer Technology Review. To obtain copies of articles and/or past issues, please contact: [e-mail].

Tony Asaro is a senior analyst for Enterprise Strategy Group (Milford, MA)

www.enterprisestrategygroup.com