Recently HDS’s Hu Yoshida wrote in his blog that he thinks the all-flash array market will go away. I very often find myself agreeing with Hu, he’s a sharp guy and I respect his point of view. As an example, later in that same blog he predicts that file sync and share will become a component of object storage; I agree with that. But as for the disappearance of the all-flash array market I have to disagree. Not only will the all-flash array market not go away but I believe that over the next 5-6 years it will become the only way that production data is stored.
To arrive at this conclusion we have essentially three trends to follow here; what is happening now, what will happen in the near future (next year) and what will happen in the more distant future (5 years +). Right now IT professionals are buying flash to solve an immediate and pressing pain point, typically caused by an under-performing application or problems specific to an environment, like virtual desktops. The flash system is bought to make this pain go away. It’s expensive, but it more often than not, accomplishes the job. Then, after the all-flash system has solved this pressing problem, IT planners look for other areas they can use it. Then, once they and their users are hooked on the performance of all-flash it spreads throughout the data center.
All-Flash at the Cost of Disk?
A question that we often ask users is, “If you could afford all-flash storage, wouldn’t you put it everywhere?” The answer, of course, is a resounding “yes”. All-flash just makes things simpler. The time to manage storage goes down significantly in an all-flash data center and the return on investment for projects like server and desktop virtualization increases as virtual machine density increases. The key question is whether you get an all-flash array for the price of a disk array. As is always the case, the answer is “it depends”.
We feel confident in saying that today you can buy an all-flash array with enterprise features and availability for about the same price as a performance focused hard disk array. “Performance focused” is the operative phrase. If you are focused on capacity and not performance, then yes, you can sell a hard disk based array for a much better price than all-flash. But that’s not comparing apples to apples. If the hard disk array is tuned for performance the all-flash array can be significantly cheaper; but there are some variables to consider.
All-flash arrays leverage a few attributes to reach their price parity with disk. First is the rapidly decreasing price per GB of the flash media itself. But as we discussed in our recent article “Flash Arrays Need High Performance Compression” all-flash array vendors also need to add storage efficiency techniques like compression and deduplication to cross the line to price parity. At this point price per GB comparisons become a little tricky because how efficient these techniques are will determine how competitive the all-flash array can be on a price per GB basis.
For example, in virtual environments (desktop or server) redundancy is quite high, so a feature like deduplication can pay big dividends. Likewise, a database environment can often benefit from compression. In other environments the payback from data efficiency is much smaller. Clearly, databases and virtual environments are mainstream so factoring some gains in price per GB for the use of compression or deduplication is fair.
The declining cost of flash media coupled with data efficiency should allow most flash arrays to be price competitive with performance focused hard disk based systems unless the storage system vendor overcharges for the flash storage. But even HP is claiming $3 per GB for all-flash arrays. If that price threshold can be crossed and the organization feels comfortable with the vendor they are getting the flash storage from, then why wouldn’t they? All-flash arrays bring high performance to a broad section of the data center. Almost as importantly, they make storage management simple and the time they save in tuning or waiting for RAID rebuilds is significant. It is for reasons like these that the all-flash array market is on such an upward trend.
In fairness much of Hu’s blog is focused on the future. Will traditional enterprise disk systems be able to catch up with flash storage and eventually be considered a better solution? After all I was very careful to discuss what we see happening now in the data center, not what the future would look like. Hu and I agree on the future, kind of. The storage system of the future will be hybrid but I don’t believe that it will have hard disks as part of that hybrid architecture. I think it will have a mixture of different types of flash storage. It might have a small SLC flash, or maybe even DRAM storage area for inbound writes, then MLC flash for the active production working set and then maybe TLC based flash for the dormant data.
The movement of data between these tiers has to be automated in some fashion but that automation does not necessarily have to happen at the storage system level. It would be easy for someone like VMware to integrate VM locality with its performance needs relative to user demands and data access patterns. This is a form of hybrid storage but it is a mixture of flash components not a mixture of flash and hard disk components. Once you and your users get accustomed to flash performance from any type of media (SLC, MLC, TLC) you are not going to want to go back to hard drive speeds, even a temporary performance drop caused by a cache or tier miss.
Where does Disk Fit In?
Over time, disk will primarily be object based storage that is used to store data that should never be on flash, large reference data or old data from production for example. Some vendors will integrate the movement of data between the all-flash array and the object storage disk tier but manual movement of this data is not out of the question. Disk, like tape, will be with us for the long term but they will serve supporting roles to an all-flash primary storage tier.