| COLUMN |
Clustered storage will win |
 |
 |
|
By Tony Asaro
26 Jun 2006 | SearchStorage.com
|
|
Most storage systems today support some level of clustering. Leading storage system vendors support dual-controller node configurations with active-active or active-passive architectures. However, storage systems that support N-way active architectures scale well beyond dual-node systems. In addition, storage systems with no theoretical limit to how large they can scale have been implemented with 80 or more controller nodes in a single cluster.
So, what does storage clustering bring to the table?
- Just-in-time scalability. Nodes can be added for additional CPU, cache memory and bandwidth. You don't have to buy based on your performance requirements down the road, but you can add resources when you need them.
- Performance protection. Storage systems that have dual nodes are in big trouble if one of the nodes fails because performance will slow down by at least 50%. With a three¬-node controller cluster, the ratio is 33%; with four nodes, it's 25%; with five nodes 20% and so on.
- Single level of management. Although you are adding more physical resources to the cluster, it is still one logical system for you to manage. It remains a single system regardless of how big it gets.
- Lower cost. Since you have already purchased the system, adding another controller node should be significantly less expensive than buying a completely new storage system. Depending on the vendor, you may not have to pay for the system software since you have already purchased it.
- Easy add-on. Adding an additional node to an existing storage cluster is much easier than implementing a whole new storage system. This will depend on the vendor and particular product, but adding a new node to a storage cluster should be an online and transparent process that requires a minimum of planning.
|
| More storage blogs |
| Read what all of our expert bloggers have to say on data protection, storage networking and more. Click here. |
|
|
|
|
However, not all clustered storage systems are created equal. Some are what I refer to as true clusters. True clusters provide a single logical storage system, so that any application can access any piece of data stored on it through any storage controller in the cluster. A true cluster also provides aggregation of all hardware resources as a single system. A true clustered storage system will greatly improve performance and provide a single level of management as more storage controllers are added to the cluster. The Isilon IQ NAS storage system is one of the best examples of a true storage cluster. Regardless of the number of physical nodes that are in an Isilon cluster, it is one logical device. This means that no matter how big the cluster gets, it remains easy to manage. When you add a new node, data is redistributed to that node, which equally load balances the entire cluster.
Another type of storage clustering, what I refer to as a management clustering, provides ease of management but not scaleable performance. The goal of management clusters is to simplify the management of the system, which is useful. However, they don't provide the linear scalable performance of a true cluster. One popular approach is to combine two controller nodes that are actually an active-active cluster and then connect this controller node pair to other pairs. There is no aggregation of performance with the other pairs.
Many SAN, NAS and CAS products provide clustering. If you think that clustered storage isn't having an impact on the industry, consider EMC Centera. It is one of the most popular clustered storage systems and has essentially created a new category in storage. A major milestone in clustered storage is the NetApp GX, a NAS storage system that supports dozens of nodes in a single cluster. NetApp jumping into the game validates clustered architectures for primary storage systems.
While this is not a complete list, some SAN clustered storage system vendors include 3ParData, Intransa, LeftHand and Pillar. Some clustered NAS system vendors include BlueArc, EMC, Exanet, Isilon, NetApp, Onstor, Panasas, Pillar, Polyserve, SGI and Terrascale. There are CAS product providers, including Archivas, ByCast, EMC, HDS, HP and Permabit, among others. There are VTL products that support clustered architectures, including Diligent, NearTek and Sepaton. Some of these products support true clustering, while others support management clusters and some a hybrid of both. The IBM SVC is a storage virtualization product that supports management clusters. EqualLogic doesn't call what they do clustering, but it does provide similar value, including scalability and ease of management.
The leading midrange SAN-based storage systems today do not support N-way clustering. Since the big storage vendors are making billions of dollars on these products, there is little motive for them to do so. But, what would happen if one of the big guys woke up and realized that they could have a real competitive advantage over the other big guys? Maybe they could break the tie and become king of the hill. Don't look now, but one of the big guys has already caught on.
About the author: Tony Asaro is the senior analyst for Enterprise Strategy Group.
|
|
|
|
