I know. I could have written from the press bull pen at the event (a very, very nice facility, by the way. Thanks, IBM!) but I was really busy learning more about IoT and hybrid clouds and fast appdev and LinuxONE and a new mainframe, the z13s. I will be publishing some more stuff about that later.
For now, my thoughts on Google’s call for drives. As reported everywhere, including ComputerWeekly, Google was asking the industry for a new disk drive. They would even help design it. It had to be bigger — that is, physically much larger than a 1 inch high 3.5 inch drive — and capable of supporting many many platters, bigger motors, and lots of read/write arms.
You know. Sort of like drives used to be.Only, they don’t seem to want a new RAMAC or DEC disk pak (see below). They want smaller spindle sizes for faster IOPS. Okay.
Oh, and they are less interested in all of that resiliency stuff. They want cheap and capacious and they will change out the disks when they fail. So, I guess the drives should be disposable too.
Readers of this blog might remember a little video we did a few years back (2011).
Notice the failure rates of drives back then — about 7% totaling 2.x million units in 2011…and those are drives that are physically smaller and designed for greater durability than Google is after.
Given that the disks being sold into the market has grown by more than 20 million units annually since then, we are probably trashing on the order of 5 million failed drives per year. That’s a lot of non-green trash for the landfills.
With Google’s lackadaisical attitude about disposable rust, and their stated intent to deploy a Petabyte of capacity per day for YouTube videos alone, we are talking about a lot of disk and a lot of disk fails. Such a strategy for mass data storage isn’t just expensive to acquire, deploy and manage, but also actually harmful to the planet — making it, I would think, unacceptable to a company that prides itself on doing no evil.
So there’s that. But there is also something else. Google has alternatives to bigger, less reliable disk. They have correctly stated that Flash is not the solution, but they have ignored the reality right under their nose.
With all of the disk infrastructure they have already deployed, they are not bound by the size of a specific drive. If they were to virtualize their disk infrastructure, they can cut volumes to whatever size they want. And with technology like DataCore’s advanced parallel I/O, they could get significantly better performance out of commodity disk than they are seeing today.
Another idea: use tape. A couple of years ago, a smart Google person, Raymond Blum, gave an outstanding talk at Fujifilm’s Global IT Executive Summit about the use of tape for data protection and disaster recovery at the company. His presentation foils are here and a video of his presentation is here. I would encourage everyone to read and watch this video because it is funny and smart.
So, Google, instead of trying to reinvent the disk drive, why not use tape in a production or archive storage role?
And that’s all I have to say about that.