For block storage, what is the difference between IOPS and throughput, and how does it affect volume choice?

Question

Accepted Answer

They measure two different dimensions of storage performance. IOPS is input output operations per second, the count of individual reads or writes the volume can do, and it is what matters for small random I/O. The classic case is a transactional database doing lots of small lookups and updates in 8 or 16 kilobyte pages scattered across the disk, where each operation is tiny but there are tens of thousands per second. Throughput is the transfer rate in megabytes per second, and it matters for large sequential I/O: streaming media, chewing through log files, running backups, where you move a lot of bytes in big contiguous chunks. The distinction matters because a workload is bound by one or the other. A database can be IOPS bound, starved on operation count while barely moving 50 MB a second, and a backup job can be throughput bound, saturating the pipe with very few operations. Optimizing the wrong dimension does nothing. So I match the volume type to the bottleneck. On AWS, gp3 gives me 3,000 baseline IOPS and 125 MB per second and I can provision beyond that. For a busy transactional database I go to io2 and provision the IOPS I actually need, say 20,000. For big sequential work I use a throughput optimized HDD like st1, which is cheap per gigabyte and gives me up to 500 MB per second. I profile first, decide whether it is IOPS bound or throughput bound, then pick, so I get the performance I need without paying for the wrong characteristic.

Answer

Is a transactional database typically IOPS-bound or throughput-bound?

IOPS-bound: it does many small, random reads/writes, so it needs high operations-per-second (provisioned-IOPS SSD), not raw MB/s. A streaming/backup job is the opposite, throughput-bound.

Answer

Why does picking the wrong volume type cost you?

You either underperform (an IOPS-bound DB on a throughput-optimized HDD starves on operations) or overpay (a sequential job on expensive provisioned IOPS you do not need). Match the volume characteristic to the workload bottleneck.

For block storage, what is the difference between IOPS and throughput, and how does it affect volume choice?

For block storage, what is the difference between IOPS and throughput, and how does it affect volume choice?

The follow-ups they ask next

What the interviewer is listening for

What sinks the answer

If you genuinely do not know

Keep going with storage

Knowing the answer is not the same as recalling it under pressure