Enterprise HDDs hit 20 TB, boost throughput in 2020

Throughput-boosting HDDs

Jason Feist

Jason Feist, managing technologist, CTO office, Seagate Technology: Hard disk drives that have dual-actuator capability will start to ramp as we move through 2020. They will lead in the cloud. We have also seen enterprise use cases where the additional 2x bandwidth is valuable for RAID rebuilds, backups and things of that nature. They are also enabling larger data lakes for analytics applications. As drives get larger, more access requests are given to them. To continue to meet service-level agreements, the hyperscale data centers need continued growth in the input/output per second metric of a hard disk drive to avoid stranded capacity.

Tom Coughlin

Tom Coughlin, president, Coughlin Associates: We're going to see 18 TB to 20 TB hard disk drives and dual-actuator hard drives hitting the market with high capacities. Dual actuators double the throughput and help with data management. For example, if a high-capacity hard drive fails, doubling the data rate halves the time to bring up a new hard drive to replace the old one. Having dual actuators helps a lot in terms of the general robustness of the storage architectures, although I don't think it's going to move them from secondary to primary storage for too many people. It gives you a performance boost, but it's not enough that you're going to push aside enterprise flash.

Phil Bullinger

Phil Bullinger, SVP and GM, Data Center, Western Digital: This is a mainstream year for HDDs built with shingled magnetic recording technology for sequentially written, read-intensive workloads. We see accelerated adoption. We're moving out of people adapting their software to the technology to using it at scale. Just-announced 18 TB to 20 TB SMR gives you a feel for the capacity efficiency gains you can get with SMR over a conventional magnetic recorded drive.

John Chen

John Chen, vice president, Trendfocus: Shingled magnetic recording will finally become mainstream in nearline HDDs later this year. Until now, major cloud vendors have struggled to make SMR drives work across a significant portion of storage workloads. However, with capacity increases slowing, the introduction of a 20 TB SMR nearline drive in 2020 should time well with the completion of host-managed SMR optimizations at hyperscale companies. SMR nearline HDDs are beginning to achieve traction at some OEM enterprise and surveillance equipment manufacturers, but the big volume inflection will occur when adoption by large cloud companies begins.

HDD trends

Chen, Trendfocus: The pace of nearline HDD capacity increases will likely slow down following the release of 18 TB conventional magnetic recording (CMR) drives in the first half of 2020. Cloud service providers are still ramping up the transition to 14 TB models, with some moving on to 16 TB, so the impact of 18 TB will be minimal until late in the year. Since all three HDD companies either have or will be switching to nine-platter HDDs to support 16 TB and 18 TB capacities, the question remains: What will come next? The addition of a tenth disk is possible. Or, some companies will plan on a wholesale transition to energy-assisted recording technologies, such as  HAMR. Both approaches will not see significant volume in 2020, and there remains a very real possibility that 18 TB nearline HDDs may be the highest capacity CMR in volume production through 2021. Our current forecast expects that full energy-assisted recording technologies will begin to see growing adoption sometime in late 2022.

Steve McDowell

Steve McDowell, senior analyst, storage and data center technologies, Moor Insights & Strategy: While SSDs have already taken over the hottest data tiers, 2020 will see flash memory start to cannibalize mechanical disk drives in midtier storage workloads. This will be enabled by the widespread availability of [quad-level cell] QLC NAND, which excels at these sorts of workloads, as well as continued pricing pressure in the overall flash market, which has reset expectations of how many gigabytes you can buy for a dollar.

Storage hardware architecture

Feist, Seagate: In 2020, you will start to see the adoption of open hardware architectures. Both instruction-set architectures as well as electronic architectures will become more readily available and collaborative. Similar to how software has repositories like GitHub where users can access, download and optimize open code, the hardware community will do the same thing. They will donate and contribute digital logic, CAD files and things of that nature so the manufacturing of that hardware can be shared and developed faster. I think it will accelerate the development of innovative technologies and hardware. Having storage devices that have open architectures and can be programmable, optimized or slightly modified can allow new business models to be created around data much faster. It will matter to enterprises because from a security point of view, anything that's an open architecture is auditable. You don't have to worry about potential proprietary influence within the architecture itself.

[Dual actuators give] you a performance boost, but it's not enough that you're going to push aside enterprise flash.

Coughlin, Coughlin Associates: We'll see RISC-V architectures showing up in storage devices. For instance, Western Digital is talking about building storage devices with RISC-V processors. It's an open storage architecture that gives you the capability of programming and building unique functions for data management, compression, deduplication and things of that sort in the storage device itself. You can't build faster and faster central processors, so people are looking a lot more at distributed computing applications. They have processors that are specialized for different functions and can perform better than a central processor. There's a general trend toward putting more intelligence in more parts of the system with specialized processors. That's tied into the computational storage that is taking off. There's an awful lot of energy and time that's spent moving data between processors and where it's stored. If you can move the processing closer to the storage, you cut down on that latency and that energy.