informa
2 min read
article

A Paradigm Spins Down

Which is better (and less cliché) than a paradigm that shifts, in my opinion. But based on public and private comments from readers, it's well past time to do away with these fault-prone spinning platters called storage arrays. Here's why.
Which is better (and less cliché) than a paradigm that shifts, in my opinion. But based on public and private comments from readers, it's well past time to do away with these fault-prone spinning platters called storage arrays. Here's why.This conversation -- that spinning drives are the data center's clunkiest dogs -- isn't a new one. But what it took to change the tenor of what's been repeated during the last five years was the voice a of a large and influential user. Specifically Google. About a year ago, the search engine giant released a study that severely undercut the performance guarantees of the major disk vendors.

Robin Harris covered this interesting issue in his Storage Mojo blog, and also included a thorough discussion of what's meant and implied in the terms mean time between failure (MTBF) and annual failure rates (AFR), two metrics that are the basis of vendor warranties.

Then, earlier this past March, some researchers from the University of Illinois did some piling on. Yes, disk failure contributes to 20% to 55% of storage subsystem failures; unfortunately, other physical interconnects like broken wires, shelf enclosure power outages, and HBA failures accounted for anywhere from 27% to 68% of problems. Ouch.

So as I look out on the horizon and see solid state, flash, even holographic storage, I wonder what it will take for the industry to wean itself off of spinning disk. Vendors have plenty invested in the systems and interfaces, not to mention a nasty dependence on the revenue stream they derive from these commodity products. Google and some academicians can't single-handedly shame makers of storage products away from disk. But they can shift the conversation so that there's discussion of amazing levels of reliability, rather than repair or replacement.

There's a paradigm that data center pros would be happy to adopt.