SPRANDOM: The Best Benchmarking Standard for Storage?

TL;DR

SPRandom is the best SSD benchmarking standard because it reaches true steady-state performance in 38 minutes versus hours with traditional methods.

Key Points

• 1.The core problem: A fresh SSD can show 3–5x inflated performance out of the box, which gradually collapses over days or weeks of real workload — making unconditioned benchmarks meaningless.
• 2.Traditional preconditioning is slow: Running a raw workload to reach steady state on a 15TB TLC drive takes 5–6 hours; the SNIA standard method takes 2–3 hours; SPRandom does it in 38 minutes.
• 3.How SPRandom works: It uses an LFSR (linear feedback shift register) to touch every LBA exactly once while simultaneously invalidating a tapering percentage of writes per zone, perfectly mimicking a drive's end-of-life state.
• 4.Key requirement: You must provide the drive's overprovisioning percentage (e.g., 7% for a 7.68TB data center drive) and run it on a raw, secure-erased device — never on a file system.
• 5.SPRandom is now built into FIO, making it accessible to anyone; the default OP setting is 15%, but users must dial it to match their actual drive spec.
• 6.Larger IO sizes speed it up further: Using 2MB random writes instead of 4K reduced the preconditioning run from 38 to ~32 minutes while producing identical steady-state results.
• 7.Manual overprovisioning trick: Limiting a 7.68TB drive to 12.8TB usable space via a DM mapper device pushed steady-state random write IOPS past 1 million — useful for architects who need more IOPS and can sacrifice capacity.
• 8.AI workloads make this critical: Inference jobs have fewer simultaneous users than databases, so a single high-latency tail IO has much greater blast radius — making true steady-state QoS measurement essential, not optional.