Perhaps I should have entitled this post “The Cyclical Nature of IT” because the same principles apply. Over time, storage moves back and forth between fully dispersed and fully centralised like a sine wave on an oscilloscope.
Look at the evidence. In the 1950’s and 1960’s there was no choice but to be centralised. All we had was mainframe and dumb disk drives. Storage needed a controller the size of a fridge to work and so was never going to be distributed to the masses any time soon.
Then in the late 1970s, in comes Al Shugart and Seagate, giving us the first drive in a format we recognise today – the ST506. As soon as that point is reached, we can now deploy our own servers with dedicated storage and so we had the distributed storage sprawl of the early 1990’s.
This quickly evolved into SAN storage, which brought us back to that centralised storage model. Since then, we’re starting to see the centralised model fall apart as storage once again goes distributed. What’s the evidence for this?
- Microsoft now recommends DAS for Exchange deployments
- PCIe SSD cards are moving storage back into the data centre
- Bespoke storage appliances (Tintri, Nutanix) are breaking the standard deployment rules
- The days of the “single large storage array” are over – SSD arrays, unified storage, dedicated NAS, object, all exist
Of course the de-centralisation of storage is done for good reasons, whether that’s cost, increased performance or otherwise gaining business advantage.
The Architects View
No doubt, we will see storage diverge further as more bespoke solutions come to the market. It’s then easy to predict that storage will once again centralise. The skill will be in making the prediction on why and when this will happen. Anyone like to place a bet?