In this chapter
We'll meet IOPS — how many separate operations a storage system can handle per second, a different question from both capacity and throughput — and use the real sequential-vs-random access distinction and a real SSD-vs-HDD gap to explain why GreenMart's checkout is slow for reasons a throughput fix would never solve.
The Problem in Real Life
"So the nightly export is a throughput problem," Mike says. "But checkout is slow too, during the day, when we're not exporting anything. What's that?"
Sarah pulls up the checkout traffic pattern — thousands of small, separate reads and writes, all day, never one big transfer. "That's not the same problem at all," she says. "That's about how many separate operations we can handle every second, not how much data moves."
Checkout isn't moving huge files. So why would it care about the same kind of speed?
Mike
Moving a Lot of Data vs. Handling a Lot of Separate Requests
IOPS measures operation count, not data volume
Many small, scattered operations (random access) stress IOPS specifically — a different bottleneck than moving one large file (throughput).
SSDs beat hard disks at IOPS by 100-1000x
No moving parts means an SSD pays almost no extra cost for random access — a mechanical hard disk pays real, physical repositioning time for every one.
IOPS
IOPS — Input/Output Operations Per Second — measures how many separate read or write operations a storage system can complete each second, regardless of how much data each one touches. It's the third and final real dimension in this Act, and it's the one GreenMart's checkout traffic actually cares about.
- Sequential vs. random access — the real distinction underneath IOPS. Reading one large file start to finish is sequential access: the storage device reads one long, continuous stretch, which is exactly the kind of workload throughput (the previous chapter) measures well. Checkout is the opposite: thousands of small, scattered reads and writes — one customer's cart here, another's inventory count there — genuinely random access, hitting different, unrelated parts of storage constantly. IOPS is what measures how well a storage system handles that second, scattered pattern.
- Why SSDs and hard disks differ so dramatically here. A spinning hard disk has to physically move a read/write head to each new location — random access forces it to keep repositioning, and that mechanical movement is genuinely slow. An SSD has no moving parts at all; reaching a random location costs barely more than reaching a sequential one. This is exactly why an SSD can be, honestly, 100 to 1,000 times better at IOPS than a spinning hard disk, even when their raw throughput numbers look more comparable.
- Why GreenMart's checkout specifically is an IOPS-bound workload. Checkout isn't moving one huge file — it's a constant stream of small, separate, unrelated operations: check inventory, reserve a unit, write an order row, update a session. None of those individually move much data, but there are a lot of them, constantly, from many customers at once. A storage system with excellent throughput but weak IOPS would still make checkout feel slow, because throughput was never the bottleneck checkout actually has.
| Storage Type | Typical Random IOPS (approximate) | Why |
|---|---|---|
| Spinning hard disk | ~100-200 IOPS | A physical read/write head has to mechanically move to each new random location |
| SATA SSD | ~10,000-100,000 IOPS | No moving parts — random and sequential access cost roughly the same |
| High-end NVMe SSD | 500,000+ IOPS | Even faster internal architecture, built specifically for heavy random access |
These are honest, real order-of-magnitude figures for typical devices, not a guarantee for any specific model — the point is the real, dramatic gap between mechanical and solid-state random access.
Three real, separate questions now have three real, separate answers: how much fits (capacity), how much can move at once (throughput), and how many separate operations can happen each second (IOPS). GreenMart's nightly export needed the second. Checkout needs the third. Neither one was ever a substitute for the other.
Key Takeaway
IOPS matters precisely when a workload is made of many small, scattered operations rather than one large transfer — and a transaction-heavy storefront like GreenMart is exactly that kind of workload, which is why its checkout can be slow for reasons a throughput upgrade would never fix.
Why This Matters
Diagnosing checkout slowness as a throughput problem — buying a faster bulk-transfer connection — would spend real money without fixing anything, because checkout was never bottlenecked on throughput. Knowing it's an IOPS problem points GreenMart at the fix that actually helps.
GreenMart now has all three real storage performance dimensions — capacity, throughput, and IOPS — each with its own real, distinct meaning and its own real diagnosis. The next chapter steps back from performance entirely, to the conceptual question this whole Act has been building toward: the three real shapes storage itself can take.