This article can also be found in the Premium Editorial Download "Storage magazine: Five cutting-edge storage technologies."
Download it now to read this article plus other related content.
Lately we're hearing about storage systems that do everything but spit-shine the dog. They are massive, low-cost, ultra-scalable, moron-ready, self-healing, butt-warming arrays of never-ending love.
I liked it when we had four types of storage: enterprise, midrange, internal and tape. We could put everything into the right category because we knew what each did--not what it claimed. Today we have 85 different classes of ILM-enabled products. We have tape that acts like disk and disk that acts like tape. We have big disk, slow disk, fast disk and cheap disk. I feel like Dr. Seuss was a Hindu and came back as a RAID controller.
Since we're now at the next great inflection point (i.e., a massively confusing time), allow my simple mind to offer an alternative to our nomenclature. Let's go back in time and refer to storage classes by what they really do, as opposed to what they could do.
First-stage storage: Traditionally called enterprise, it has the highest fault tolerance, and cost, very high performance (though perhaps not the highest) and can connect to open systems and mainframes. Might be monolithic, might be modular--I don't care anymore. It scales predictably in all three dimensions--capacity, I/O and throughput.
Second-stage storage: traditionally called midrange. Modular, probably the fastest on a controller vs. controller basis for I/O or throughput. It's lower cost because it has less stuff. This is where we keep just as much, if not more, capacity as first stage, but for longer periods of time because of the cost. ILM life = 18 months. Capacity = 3X.
Third-stage storage: bigger bulk storage, cheaper still. Less bells and whistles, and really idiot-proof to manage. One person manages a zillion terabytes. It grows by waving your hand over it. We keep stuff around for a long time here. This stuff includes the data protection classes--the virtual tape libraries that cache between stage one and two to tape at the end of the line. It's also where we keep our continuous-capture data. In short, this stage is at the very least becoming the new initial backup target. Our backed-up data stays here for a long time, so we never recover from tape unless all hell breaks loose--like the Red Sox winning the series. ILM life = 36 months. Capacity = 15X.
Fourth-stage storage: bulk storage. Massive disk archive storage that stays sort of online forever. This is the bit bucket where we never delete anything, ever. Capacity = 100X.
Fifth-stage storage: Tape/optical/whatever removable permanent media. Massive terabyte/cartridge/platter. Write it once, whenever, because it doesn't matter about performance any more, make it smart so we can find things if we absolutely have too--and shove it in a bunker.
There are more intricacies, of course, as there always are. But we need to figure out how to map technologies to all this ILM garf, so that some of this stuff starts to make sense. We need to stop positioning gear that's wonderful for certain tasks as the end-all for everything. Albert Einstein could never hit Pedro Martinez, and I don't think Albert Belle could even spell relativity.
This was first published in October 2004