Guy Bunker

SANs, NAS, iSCSI, virtualization, in-band, out-of-band, the terminology seems never ending when it comes to storage and what's worse, no one will tell you what's best. Unfortunately, it's not that simple. The advent of SANs and the introduction of new technology has increased the number of options available, but there are no clear guidelines as to which one to use and when. There isn't a silver bullet or golden configuration that is good for everyone; the solution has to be tailored to the specific environment. But all is not lost. There has been a lot written about storage and storage architectures, and if all else fails, look at what you are trying to achieve and how much money you have to spend.

While it is widely thought that SANs are for big enterprises and NAS for smaller ones, this is not true. Most enterprises, whether big or small, now have NAS servers and many are using them for more than just file serving. The cost of SANs has fallen such that they are now a very real prospect for smaller organizations that want to take advantage of improved connectivity and performance to utilize with technologies such as third-party copy and clustered file systems. So it is the applications and the business requirements that should drive the architecture, not the "latest and greatest" technology or the cheapest solution. Storage is not just about the online disk. Backup (which now might be to disk before going to tape), disaster recovery, and legislative compliance all have their part to play. Without a big picture of what needs to be achieved (from the business perspective) the decisions made will be insufficient. Another factor to include is storage growth. If the space required in 12 months is 100% more than you have today, will that influence your architecture decision? What happens if it is 1000% in three years? How long do you plan to remain with the architecture that has been defined? The immediate logical conclusion is to go for the biggest you can buy - now. But we know this is not a pragmatic business decision; the architecture should be designed so that it can be grown - and this might mean starting with NAS and expanding into a SAN just as much as starting with SAN and acquiring a NAS solution later.

Utility computing is a trend we are hearing a great deal about, with many vendors touting it as the next big thing. When it comes to storage, applying utility computing principles and creating a storage utility is a great place to start. By using storage virtualization tools storage can be pooled and then provisioned when required; by having it attached to a SAN it can be allocated to any server that needs it. Additional functionality allows file systems to be grown automatically without the need to take the application using it down. Business reporting tools enable departments (or lines of business) to see how much storage they are using. The IT organization can then choose to apply costs to the storage and could present each business with a bill (a.k.a., chargeback) if they so wished. More often than not it is the insight into costs that is useful, and it can be an invaluable guide as to where best to invest money in IT to get the greatest return for the business. In addition, utility computing is all about improving efficiency through best practice and automation. Again, storage is a great place to begin and putting in some best practices and simple automation - e.g., increasing space on servers when they are running out - can save a business a great deal of money, no matter what its size.

The grid is also seen as the next big thing and again, storage is a key component of a grid architecture. However, most grid applications need a large amount of space to store data centrally; that data is then farmed out and generally processed in memory within the grid so the actual storage requirements are virtually nonexistent on the fringe nodes. For the main central storage, ensuring that the application serving out the data is highly available and that the data is sufficiently protected, i.e., backed up or replicated, is generally adequate. Outside of storage, a general comparison of grid versus utility computing is interesting because, while both have very different applications running on the architecture and so from 30,000 feet look very different, from the ground level there are many similarities: what is being used, how much it is being used, and if it can be used more - to improve either efficiency and/or utilization.