There is little doubt that solid state disks have become a disruptive force for datacenter storage and have bright future. Their steady rise, either in all-flash arrays or disk-flash hybrid systems, is being driven by declining prices for NAND and the rapid adoption of NVMe.
This is leading some industry watchers to ponder the demise the spinning disk. But this device still has a few tricks up its sleeve and is continuing to evolve right alongside its solid state competition.
As we pointed out just last month, while SSDs in the datacenter are growing at double-digit rates, so are HDDs, albeit at a slightly lower CAGR. And given that the deployed base of hard disks is much bigger, they represent the larger market opportunity.
IDC forecasts that by 2025 175 zettabytes of data will be created, representing more than a 5x increase over the volume created in 2018. The economics of storage dictate that not all of this data will be saved, but this level of data creation will still trickle up into an expanding pie for SSDs, HDDs, and tape.
As long as spinning disks maintain their price-capacity advantage relative to solid state drives, they will continue to dominate datacenter storage for the foreseeable future. At our Next IO Platform event, Western Digital Marketing Vice President Dean Amini outlined some of the more promising technologies that they are developing to make this possible. He pointed out that these efforts are being performed in conjunction with the demands and preferences of some of their biggest clients, especially hyperscalers.
As we reported before, the search giant has its own ideas about how to make hard drives more cost-effective. Their interest in encouraging component suppliers like Western Digital to adopt some of them stems from the belief that rate of increase in capacity-per-dollar for HDDs and SDDs can be maintained. Amini believes that as well, stating that “if you want to store massive amount of data, you need to be able to store it on something that provides the right economics.”
One of the most promising technologies to increase disk capacities is energy-assisted magnetic recording. It comes in two forms: heat-assisted magnetic recording (HAMR) and microwave-assisted magnetic (MAMR). Western Digital has explored both methods, although Amini didn’t delve into that particular dichotomy. (Based on previous statements by the company, it is thought to be exclusively focused on the MAMR technology at this point, while Seagate appears to have opted for HAMR.). Both versions, however, promise a significant increase in areal density over the traditional perpendicular magnetic recording (PMR) technology used in hard disks today.
Another way to increase density is with shingled magnetic recording (SMR), a method in which a portion of a track overlaps an adjacent track, much like roof shingles. How it does this is somewhat mysterious, but it takes advantage of the fact that the overlapped tracks can be trimmed without losing data integrity. Seagate has had SMR devices since 2014, claiming 25 percent more capacity per disk. Amini explained that this technology is better suited to sequential I/O, implying there may be some sort of performance penalty with more random access patterns.
It’s worth noting that last month Western Digital announced it was sampling a 20TB SMR disk drive for select datacenter customers by the end of the year, with volume shipments expected in the first half of 2020. The company predicts that half of the HDDs its sells in 2023 will use SMR technology. A new 18TB disk, which is also being sampled to customers this year, incorporates its energy-assisted magnetic recording technology, which again, is assumed to be based on MAMR.
Besides SMR and energy-assisted recording, Amini also mentioned they are looking into allowing the host to perform intelligent data placement on the disk. The idea here is that different types of workloads have different I/O access patterns, which can be accelerated by optimizing their placement on the spinning media.
Other ideas involve additional hardware, such as building disks with larger diameters or adding more actuators. The first case would undoubtably increase unit costs but could potentially increase capacity per dollar. Multiple actuators would be more expensive as well but promise faster read access.
“We and our customers are exploring all of these,” said Amini, “and when they make sense economically and in terms of infrastructure support, that’s when we will see them in the market.”