The future Role Of Spinning Tin in the Data Center
My friend John Obeto ( @JohnObeto ) asked me for my opinion on what may be the future of the “Spinning disc” in three categorical use cases within IT infrastructure 1) DataCenter 2) Local or Small Server and 3) in the Desktop. I enjoy these conversations when it’s purely gazing into the crystal ball, because I’m not necessarily any more clued into manufacturing predictives than anyone else. I am, however an employee for an organization that leverages the power of solid state disc for IO.
As an employee for Nexenta Systems, we use both spinning disc and solid-state disc in a tiered methodology in our common methodology, as ZFS has most regularly implemented. Of course, the flexibility of ZFS doesn’t lock in to a specific configuration, by that I mean, one may configure all SSD for pure IO configurations, or even all spinning disc in a use-case that needs very little in the way of IO. The point being, I configure appliances on a daily basis that lean on the power of both spinning disc and solid state regularly and have developed a familiarity with the SSD industry as it is, and as such, feel that maybe I have some insight into where we may be going into the future. That being said, let’s dive in-
In the data center, I am already seeing many organizations who’ve no desire to invest at all in spinning disc, even in their low latency use-cases. They see spinning disc as dead technology, and don’t understand even in slow disc cases why anyone would invest in ancient tech. On the other hand, and still today, the cost differential presents quite a barrier. This difference will continue to shrink, in my opinion, to where I believe even the smallest organizations will see the benefits outweighing the negatives in a situation like expressed above. Where pennies per gigabyte will be close enough, and mean-time-between-failure rates will surpass spinning disc, so as to make the logic behind spinning disc, as reasonable as, for example, cranks to start a car, or party line telephone calls.
Local servers may be a little slower to adopt this, but I do feel that we’re going the same way. And in the same manner, I’m seeing many servers adapting their internal storage to a solid-state form. Local storage may be, at the moment an admittedly more expensive and smaller, but this is changing. The size of discs is growing, and though prices are still relatively high, only in comparison to today’s disc prices. They’re dropping daily, and at this level the cost really shouldn’t be the issue. In fact, one of the reasons we used to use local disc was raid level, not for redundancy only, but with the high IO levels of SSD, certainly, we’re not reliant on spindle count. Maybe rather than a 4 disc raid 6, we’d only require a 2 disc mirror for redundancy, thus mitigating some of the cost, as well.
And, in the final use-case, I’m really hard-pressed to satisfy my desire for ample local disc in a desktop workstation with SSD. Were I to be buying a non-laptop today, and to be fair, I work about 90% of the time from a laptop, I would build out with a couple of largish SSD’s, and then connect for the lion’s share of my storage to my NAS (which to be fair is a NexentaStor device). I still feel the future is the same, and well defined for spinning disc regarding desktop machines. Only the dinosaurs will have spinning discs. We will likely regard these machines as we do with nostalgia, our old NetWare boxes, or our first PC XT’s.
So, what happens in the interim? Will there be a place for the “Hybrid” disc? We’ve seen a place for these in laptops, where 2.5” discs which have been able to be augmented by a small amount of RAM which facilitates a pseudo disc tier within the physical disc architecture, and creates a much faster disc relationship. But, will we be able to realize this in larger sizes in all architectures before the full SSD overtakes the spinning disc entirely? My belief is that we will not.
Another consideration to lay thought towards is that of the ultra-fast PCIe ram card, like today’s FusionIO or Intel910 card. These remove the card from the SAS or SATA bus, to the PCIe socket to further improve IO throughput. They tend to be placed in appliance only Flash devices, or augment individual solid-state local device appliances. They’re wonderful, however, they tend to be more expensive on a dollar-per-gigabyte level than SSD disc in equivalent sizes. I am not sure what the future brings for these devices, but I anticipate that we’ll see parallel growth in these markets.
Ultimately, my prediction is that we’re at a tipping point, and that the industry is turning towards SSD in a very big way. Commodity disc and enterprise class regardless, will, I think, overtake spinning disc entirely, and I actually think that we’ll be seeing that sooner than the 5 year mark.