From Bottleneck to Breakthrough: Solving the SSD Preconditioning Challenge

There are specific benchmarks that make an engineer distinguished: contributions to a standards-making organization or a certain number of patents, for example. Steven Sprouse has those in spades, but he defines the role not through accolades or acclaim, but through impact and collaboration.

"One of the things I do is look beyond my projects, to pull ideas and technologies from a wide variety of places and people to solve problems throughout the company," he said. Sprouse, a distinguished engineer in Sandisk's enterprise SSD architecture group, did this for decades: listen, observe, collaborate, solve.

That's how he got involved in solving the problem of preconditioning SSDs, by listening to his colleagues and lending his wealth of knowledge to the solution. His approach will save engineers countless hours, but for Sprouse, the true success is contributing to the broader community and advancing an entire industry.

Preconditioned

Before Sandisk ships an Enterprise SSD to a customer, it undergoes a lengthy test and qualification process. One of the most common tests requires preconditioning drives, a process that puts them in a state resembling their real-world behavior. To do this, data is repeatedly saved, copied, and deleted until steady-state performance is consistently achieved. From there, the drive is ready for the real-world workloads that testing demands.

Historically, this has been done with a two-phase approach: a sequential write to every address, and a random write to redistribute overprovisioning (extra capacity in the drive used for moving data around) across the drive, achieving steady-state performance. When drives were small, under 16TB, this method worked just fine, but as capacity has climbed, preconditioning time has increased dramatically. While capacities are higher, the maximum write rate of the drives has kept pace, resulting in a significant increase in preconditioning time.

"During the development process, an important metric is the 'run, break, fix' loop time. How many times a day can we run a test, see where it breaks, and then fix the problem?" said Sprouse. Smaller drives can get through a loop in a couple of hours, but that's no longer the case for the highest capacity drives in demand.

"Now, pre-conditioning takes 12, sometimes 24 hours and longer, so we couldn't do even a single run, break, and fix loop in a day. That's really where we started to dig into the problem", he said.

Simple, efficient, and fast as hell

While mulling over this challenge with colleagues, Sprouse recalled that over a decade ago, another Sandisk engineer with a strong math background derived a formula describing how overprovisioning was distributed within a drive after preconditioning. Sprouse and a few other engineers had the initial idea to apply this formula to achieve rapid preconditioning a decade ago. However,it wasn't deemed necessary at the time because preconditioning times were still relatively short. Fast-forward to the current demand for very high-capacity drives, and preconditioning time suddenly becomes a major problem.

SPRandom (SanDisk Pseudo-Random), the all-in-one algorithm, divides the drive into several overlapping sections. The amount of overlap corresponds to the amount of overprovisioning expected by the math formula for each section of the drive; as the physical addresses increase, the amount of overprovisioning decreases. The algorithm also ensures that each address is written at least once in a pseudo-random sequence.

Through this methodology, SPRandom achieves all the goals of preconditioning. It writes to every logical address, distributes overprovisioning across the drive, and achieves steady-state performance in a single physical drive write. What took old methods some 160 hours takes SPRandom about six and a half hours, according to a technical presentation Sprouse gave at an OCP tech talk — a game changer.

Sharing innovations

It would be sensible to patent the idea and use it as a competitive advantage for the company. But, as Sprouse put it, there's often more to gain from sharing knowledge than hoarding it.

"If we want to lead the industry, we have to be a part of the larger community and its discourse," he said—a part of, not apart from. Gatekeeping the technology may help Sandisk and its customers, sure, but sharing it with the broader community has benefits, too.

"I've had people ask me why we aren't keeping this a secret, but I just think there's more to gain by sharing it with the wider industry," Sprouse said. "FIO is an open-source tool that we all benefit immensely from. Why not make it better?"

To bring this innovation to the masses, Sprouse has worked with a team to integrate SPRandom into the open-source FIO platform, enabling the entire industry to accelerate preconditioning. If it's easier to precondition and test larger, in-demand drives, that's good for vendors and customers alike. A rising tide lifts all boats, after all.

"We believe, long-term, this will help us and the industry. This is one small part of getting higher-capacity drives adopted sooner," said Sprouse.

A global community of innovation

Open-sourcing SPRandom exemplifies Sprouse's approach to his work. It's not about patents and prestige for him; it's about tackling challenges with like-minded peers, solving them with engineering excellence, and contributing that knowledge to a broader scientific community. Since becoming an independent company, Sandisk has been encouraging more of its engineers to think like Sprouse.

"Since the split, we've been encouraged to focus on the broader community a bit more, to participate in the industry groups. That's where the exciting problems are: at the industry level," he said.

Sprouse's work proves that leadership is about more than gaining an advantage and holding on for dear life. It's about creating new, novel ideas, implementing them in creative ways, and pushing the boundaries of what's possible for everyone.

"There are so many opportunities in front of us to make a difference in the industry. To have people at all levels of the company thinking about the larger ecosystem, our role in that large field, it's an incredible opportunity for us engineers, yes, but for the company, too."