How Much Slower a Disk Seek Is Than a RAM Read
A main-memory reference takes ~100 nanoseconds. A spinning-disk seek takes ~10 milliseconds.
How many times slower is the disk seek — and what does that ratio look like at human scale?
Place an SSD random read on the same ladder.
C) 100,000× — five orders of magnitude between two things casually called "storage."
10 ms ÷ 100 ns = 10⁷ ns ÷ 10² ns = 100,000×.
At human scale: if a RAM reference took 1 second, a disk seek would take ~28 hours. That's the difference between glancing at a sticky note and mailing a letter and waiting for the reply.
The ladder (the classic "latency numbers every programmer should know"):
| Operation | Latency | vs RAM |
|---|---|---|
| Main memory reference | ~100 ns | 1× |
| SSD random read | ~100 µs | ~1,000× |
| Spinning-disk seek | ~10 ms | ~100,000× |
| Same-datacenter round trip | ~0.5 ms | ~5,000× |
This one ratio explains half of systems design: why databases obsess over keeping working sets and indexes in RAM, why one disk seek per request caps a drive at ~100 requests/second, why caches exist at every layer, and why B-trees are shaped to minimize seeks rather than comparisons.
The rule of thumb to say out loud: RAM ≈ 100 ns, SSD ≈ 100 µs, disk seek ≈ 10 ms — each storage tier is roughly 100–1,000× the one above it, so design to stay as high on the ladder as possible.
Share this question