Kit Wai Chan - Fotolia
Solid-state drives that use dense QLC NAND are becoming an intriguing alternative to spinning disk because they can reduce costs over flash options that store fewer bits per cell.
Pure Storage and Vast Data -- two vendors that make arrays with quad-level cell (QLC) flash -- went so far as to unequivocally predict that QLC SSDs would replace hard disk drives (HDDs) during a panel discussion at last week's Flash Memory Summit.
But most storage experts agree that, much like age-old tape media, cheaper HDDs won't disappear any time soon -- especially with some of the world's biggest storage users still relying on them to keep costs down.
Los Alamos National Laboratory (LANL), for instance, uses HDDs to store close to 300 petabytes (PB) of cool data that would not benefit from the higher IOPS that flash drives can deliver. Gary Grider, high-performance computing division leader at LANL, said he does not see QLC SSDs replacing disk at the lab.
SSD cost differential
"There is still a 4x difference in cost per capacity," Grider said. "Disks can produce reasonable bandwidth per dollar if used in mass parallel, which we have been doing for decades. So, for applications that require only bandwidth, with no latency requirements, disk is the answer -- and will be for years, we think."
Grider said LANL uses the HDDs for backup, archive, transfer and staging applications, as well as slow offline indexing of cold data. He estimated that the lab also stores about 500 PB of cold data on tape and 6 PB of recently generated data on SSDs.
"The flash is growing faster by percentage than the disk, which is growing faster than the tape," Grider said. Although LANL doesn't have much QLC flash, Grider said he expects the lab's QLC SSD numbers to change "soon and fast."
Like all NAND flash, QLC technology can speed access to data over the mechanical HDDs that account for the bulk of the world's shipped storage capacity. But to realize QLC's cost benefits, storage manufacturers must overcome a number of technical challenges, including its lower endurance, performance and throughput compared with single-, multi- and triple-level cell flash that store one, two and three bits per cell respectively.
Addressing QLC challenges
Pure Storage, for instance, built proprietary DirectFlash modules to connect the raw QLC flash directly to its storage system via NVMe and facilitate the data placement, higher throughput and lower latency necessary for its FlashArray//C to handle performance- and capacity-oriented workloads. Shawn Rosemarin, vice president of worldwide systems engineering at Pure, said during the Flash Memory Summit panel discussion that latency would be the No. 1 reason that users replace HDDs with QLC flash.
Shawn RosemarinVice president of worldwide systems engineering, Pure Storage
"There's no doubt about it. Spinning disk is on the declining curve where it will only exist for the lowest-value data," Rosemarin said.
Rosemarin said some vendors front end QLC with TLC flash, storage class memory or DRAM to address caching and performance issues, but they run the risk of scaling problems and destroying the cost advantage that the denser flash technology can bring.
"We had to launch a whole new architecture with FlashArray//C to optimize and run QLC," Rosemarin said. "Otherwise, you're very quickly going to get in a position where you're going to tell clients it doesn't make sense to use QLC because [the] architecture can't do it cost-efficiently."
Vast Data's Universal Storage uses Intel Optane SSDs, built on faster, more costly 3D XPoint technology, to buffer writes, store metadata and improve latency and endurance. But Jeff Denworth, co-founder and chief marketing officer at the startup, said the system brings cost savings over alternatives through better longevity and data-reduction code, for starters.
"We ask customers all the time, 'If you had the choice, would you buy a hard drive-based system, if cost wasn't the only issue?' And not a single customer has ever said, 'Yeah, give me spinning rust,'" Denworth said.
Denser NAND flash chip technology isn't the only innovation that could help to drive down costs of QLC flash. Roger Peene, a vice president in Micron's storage business unit, spotlighted the company's latest 176-layer 3D NAND that can also boost density and lower costs. QLC could also benefit at the drive level from new technologies such as Zoned Namespaces, Peene said.
Keeping HDDs competitive
But Ken Steinhardt, field CTO at Infinidat, argued that users don't care which technology they use so long as it addresses their business needs on price, capacity, performance, reliability, stability or whatever criteria they set. Steinhardt sees a bright future for QLC, follow-on penta-level cell (PLC) flash, storage class memory, DRAM and HDDs. He noted innovative efforts to keep spinning disks competitive, including shingled, heat-assisted and microwave-assisted magnetic recording as well as dual-actuator HDD technology.
Infinidat takes a technology-agnostic approach and uses flash and faster DRAM to cache data and conventional HDDs to store data persistently on the back end. Steinhardt pointed out that, although QLC SSDs might have eclipsed the cost of traditional 15,000 RPM and 10,000 RPM HDDs, a recent Gartner forecast shows the price-per-GB delta between nearline enterprise HDDs and server-class SSDs is about 9x.
"And, curiously, they're forecasting that all the way through 2024, it's still going to be [at least] 9x, so to me, the issue isn't QLC or hard disks," Steinhardt said. "I believe there's going to be a place for both."
Thomas Isakovich, CEO at Nimbus Data, whose storage portfolio includes an all-flash array platform and a 64 TB QLC SSD, said hyperscalers "get bent out of shape over fractions of a penny," so it won't be a "winner-take-all game." Data reduction technology -- "supposedly the great savior in making flash cheaper" -- doesn't do well on emerging data types such as unstructured content and media, he said. Plus, NAND flash supplies are insufficient to store all the data in the world.
"It's sort of like when people say, 'Hey, the whole country needs to run on solar and wind.' It's not possible to power this country on solar and wind," Isakovich said.
Don Jeanette, a vice president at Trendfocus, said NAND manufacturers produced 8 exabytes of QLC flash in the third quarter compared with about 150 exabytes of nearline HDDs.
"Think about what type of QLC ramp it would take to even consider displacing the type of storage capacity being shipped by nearline HDDs," Jeanette said.
Long tail for HDDs
One point of agreement among most storage experts is that HDDs will have a long tail. Pure Storage's Rosemarin predicted the crossover from QLC flash to disk would happen within three years. IDC forecasts QLC SSDs would not replace HDDs through at least 2025. Isakovich predicts that HDDs will stick around for at least a decade, if not two.
Tom Coughlin, president of Coughlin Associates, said QLC SSDs would have the greatest impact on client storage and as a new tier of capacity storage at the enterprise and hyperscale level, as performance-based applications continue to move from HDDs to flash.
"Just as cheap HDDs have not eliminated tape, cheap QLC will likely not eliminate HDDs," said Marc Staimer, president of Dragon Slayer Consulting. "Right now, the cost/performance of QLC SSDs is better than HDDs. But if performance is not an issue for the use case, HDDs are a better fit. When QLC prices become equivalent to HDDs, then there are few use cases, if any, that favor HDDs."