I would suggest that perhaps it is time for us to retire RAID in its current form and bring in an entirely different paradigm into storage systems for all sizes, from consumer electronics to the largest arrays. I bet most of you are already wondering if I have gone nuts. Why, you ask, would we want to eliminate the best thing that has happened to storage from a data protection and data integrity point of view? There are many reasons, as I explain below.
First, let's start with the RAID issues of today.
The list is long, but I will use, say, five examples, to illustrate the point.
First, ask yourself why I need the same size drives to create a RAID group. Since drive capacity is constantly getting larger, why can't I simply add a drive that is most readily available in the market and not lose the excess capacity? Second, why should I need to decide right up front that I need RAID-1 or RAID-5? Can the system not decide, based on the number of drives available, how to best protect the data? Third, why should I have to go through the miserable process of migrating data, changing the RAID setup, say, from RAID-1 to RAID-5, and then migrating the data back? Fourth, why is it so painful for me to change from 3+1 to 4+1 in RAID-5? Fifth, can someone please tell me why I should continue to live with rebuild times measured in tens to hundreds of hours after I insert a drive in a RAID-5 group? Of course, we all know that during the time the rebuilds are happening, the storage system is vulnerable to data loss if another drive fails.
Actually, now that I have got you going, I bet if I asked you to add to the above list you would come up with a dozen additional reasons why the current state of RAID is untenable. The problem is precisely that. That no one has started with a clean slate of innovation for more than two decades. Instead, as an industry, what we have done is to constantly improve RAID. We use faster processors for XOR, the logical function involved in calculating parity in a RAID-3, 4, 5 and 6. We have added automated methods of migrating from RAID-1 to RAID-5 (actually, even this is rare today). We allow a hot spare to automatically replace the failed drive so a rebuild can start without operator intervention, etc. All good things, but all designed to live within the existing paradigm.
I am aware of a few startups that are effectively going to change all this. They are not far from delivering their first product. Granted, the first product is likely designed for the smaller application environment, even the consumer market. If you think about it that is where the biggest need is. That is where you expect the user not to know anything about RAID. Data just needs to be protected in the best possible fashion, given the drives available -- all behind the scenes, and all without any fanfare.
Start with a single drive and when a second drive is added, mirroring is automatically used. As you add drives, even drives of unequal size, the system adjusts itself for the most efficient use of all these drives while continuing to guarantee protection against drive failure. When the system begins to fill up, it will inform you to add a drive of any size you desire. When the system maxes out, just replace the smallest drive with a larger one. When you do, everything happens behind the scenes to protect data in the best possible fashion, given the number and the size of the drives. The user doesn't have to know anything about RAID. All he does is take the action indicated without complex management consoles or configuration. While it all appears simple to the user, the amount of technology to make this happen is not trivial. That's why it hasn't happened until now.
To be fair, I think the basic principles of this innovative thinking started a while back with the new storage system players, such as Compellent and 3PAR and Pillar and others. They have indeed made RAID more flexible and easier to use. They have added the concept of virtualization beyond RAID. But to me, the "clean slate" startup in this area is Trusted Data. Their product is due in the market early next year, and it simply establishes a new standard for data protection within a storage system.
RAID is indeed long in the tooth. It has served the industry well for two decades. Patterson could never have foreseen the impact his white paper would have on the industry. And, of course, nothing dramatic will happen for another three years. But I fully expect that over the next few years we will not have to "configure a RAID set" or buy the same size disks just to keep the RAID set happy. In fact, I would go as far as to say that young IT folks a decade from now may not even know what RAID is. Worse, if one of you more experienced guys used the word RAID they will run, thinking they were being raided by aliens -- or the police.
About the author: Arun Taneja is the founder and consulting analyst for the Taneja Group. Taneja writes columns and answers questions about data management and related topics.