Become a TechRadar Insider
- Sandisk reduced SSD pre-conditioning times from days to only several hours
- SPRandom achieves steady state performance using only one complete drive write
- Open sourcing SPRandom gives hyperscalers faster deployment options
Sandisk has open sourced a key algorithm which can dramatically speed up the pre-conditioning process for solid-state drives (SSDs).
SanDisk Pseudo Random (or SPRandom) reduces the time required to prepare a drive for steady-state operation from over 160 hours to roughly 6.5 hours.
Traditional pre-conditioning methods require writing data two or more times the drive's total capacity using sequential and random operations. But the new algorithm writes to every logical address just once, completing the entire process in less than five percent of the original time.
SSD pre-conditioning matters for data centers and AI workloads
Fresh out of the box SSDs show variable performance, until they go through a process called pre-conditioning that stabilizes their behaviour.
During this process, the controller fills the drive, starts garbage collection, manages wear levelling, and distributes over-provisioned blocks across the storage area.
This background activity must reach a steady state before the drive's IO performance becomes predictable and reliable for production use.
Traditional pre-conditioning of a 128 TB drive takes over 160 hours, or nearly seven days, but Sandisk's SPRandom completes the same task in just 6.5 hours.
A 256 TB drive requires up to 250 hours, roughly 10.5 days, using conventional methods, while SPRandom again finishes in only 6.5 hours.
These figures imply that Sandisk's SPRandom reduced the pre-conditioning time by between 95% and 97.4%.
The SPRandom algorithm divides the drive into overlapping sections, with the amount of overlap corresponding to the over-provisioning expected for each section.
As the physical addresses increase, the amount of over-provisioning gradually decreases across the drive.
The math behind SPRandom computes how over-provisioning gets distributed after pre-conditioning, ensuring steady-state performance is achieved in a single physical drive write pass.
Sandisk has released the SPRandom code as an extension to the FIO tool, which stands for Flexible IO Tester, making it freely available to the entire storage industry.
This matters for AI but not for your gaming PC
Hyperscale data centers and AI infrastructure operators buy SSDs in massive volumes and need them ready for production as quickly as possible.
Shaving more than 150 hours off the pre-conditioning time per drive translates directly into faster deployment and lower operational costs for cloud providers.
AI workloads are particularly sensitive to storage performance variability because training and inference tasks demand consistent IO behaviour across thousands of drives working in parallel.
But, for a typical gaming PC with a single drive, spending an extra week pre-conditioning a new SSD is simply not relevant to the end user experience.
The technology delivers enormous value at scale, but the average consumer will never notice the difference.
Sandisk's decision to open source the algorithm is genuinely generous, but the beneficiaries are data center operators with massive storage arrays.
Simply put, the math is clever, the time savings are real, and the impact on AI infrastructure could be substantial.
However, your gaming PC will keep working exactly as it always has, and that is precisely the point.
Via Blocksandfiles
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.
