SSDs have been the subject of hype, hype and more hype. Any of us who have used them in our personal computers know the benefits SSDs bring, but personal experiences and hype alone don't explain the robustness of enterprise flash adoption.
Flash has a lot of naysayers. Any article on The Register about flash inevitably invites at last one crank commenter who feels the need to tell the world how flash is just too new and they wouldn't be caught dead deploying it in their data centre.
As someone who uses flash in production, I wonder when (or if) the naysayers stopped using mercury delay lines and magnetic core memory. Or do they just etch the 1s and 0s directly onto recording wire manually with an electron microscope?
It is easy to make fun; I have taken the plunge and run flash under real world conditions for years. I have burned out flash drives well before their supposed lifetimes and seen flash drives that seem to last forever. I have a feel for what is going to wear flash out and what isn't, but despite having run up against its very real limitations, I certainly wouldn't go back to magnetics.
But flash does have technical limitations and they are easily magnified by those who don't know what they are doing. Flash is faster than traditional magnetic disk, but it is not a like-for-like replacement and it won’t kill traditional disk.
With that in mind, let's take a look at where flash is useful and where it is not, and what to bear in mind as you begin to move towards enterprise flash.
Get in line
This biggest issue is that there is only so much flash to go around. Two things form a hard limit: physics and economics.
The physics of flash is a little complicated, but suffice to say that the more we shrink the manufacturing process for SSDs, the lower the write endurance of each cell becomes.
We are almost out of road with regards to process shrinks: by 2020 we probably won't be able to shrink the transistors any more and still have even remotely useful write life. 3D flash will allow us to stack transistors on top of one another and this will buy us some capacity increases for a few more years but that is a band-aid, not a cure.
Even as we approach the physical limits of our ability to shrink flash processes (and thus keep the $/GB dropping) the industry is still hyping flash like there's no tomorrow. And that brings us smack dab into the next hard limit on flash: economics.
Flash is awesome and it is in demand. There are only so many fabs in the world that can make flash, and they are the same fabs that could make the also much in-demand RAM chips. Every flash chip we make is a RAM chip we don't make, and planetary demand exceeds supply. This keeps prices high for both flash and RAM.
Fabs are outrageously expensive to build and they require expertise and equipment that is not easy to come by. Globally, we are building not just the fabs, but all the other elements of the supply chain as quickly as money, people, space and equipment can be found. Despite this, flash does not look set ever to eradicate magnetic disk.
The simple truth is that global growth in data demand is so high that we physically couldn't put our planetary data demand all in flash, even if we wanted to. And we never will, as flash will hit the wall before we ever get close, forcing us on to post flash technologies.
Flash then – especially for high data usage enterprises – needs to be approached with careful consideration. There is only so much to go around, so don't use it where it doesn't make sense.