EBS volume-type questions look like storage trivia and are really a requirement-matching drill. Every EBS scenario names a bottleneck — random IOPS, sequential throughput, cost per gigabyte, or durability — and the volume family splits cleanly along those lines. Read the requirement word first, and the four options sort themselves. The diagram above lays the family out the way the exam reasons about it: SSD versus HDD across the top, and instance store set apart because it is not EBS at all.
SSD is for IOPS and latency
The two SSD types answer questions about random IOPS and low latency. gp3 is the general-purpose baseline, and its defining trait on the exam is that you provision IOPS and throughput independently of volume size. That is the fix whenever a stem describes a small volume that still needs solid IOPS — you no longer over-buy capacity to unlock performance the way gp2 forced you to. Reach past gp3 to io2 (and io2 Block Express) only when the numbers get extreme: very high sustained IOPS on a single volume, sub-millisecond latency, and 99.999% durability for a transactional database. If the requirement does not push into that territory, gp3 is the cost-optimised default, and picking io2 anyway is the trap.
HDD is for throughput and cheap capacity
The two HDD types answer questions about sequential throughput and cost per gigabyte — never about low-latency random access. st1 is throughput-optimised for large, frequently accessed sequential scans, the kind a big-data cluster runs. sc1 is cold HDD: the lowest cost per gigabyte, built for infrequently accessed data. The line between them is access frequency — frequent-and-throughput-bound is st1, idle-and-cheap is sc1. And the moment a stem mentions random small reads and writes or consistent low latency, both HDDs are wrong on principle, however tempting the price looks.
Instance store is the ephemeral outlier
Instance store is local disk on the EC2 host — not an EBS volume — and it earns its place in these questions by offering the highest local I/O at no separate storage charge. The catch is that it is ephemeral: the data is gone when the instance stops or terminates. So it is the right answer only when the data is rebuildable — a cache tier repopulating from a database, for instance — and the goal is raw I/O rather than persistence. If the workload needs the data to survive a stop, instance store is disqualified no matter how fast it is.
The requirement word that decides it
- Small volume, moderate IOPS, sane cost → gp3.
- Very high sustained IOPS, sub-ms latency, top durability → io2 / io2 Block Express.
- Frequently accessed sequential throughput on a budget → st1.
- Infrequently accessed, lowest cost per gigabyte → sc1.
- Highest local I/O on rebuildable data → instance store (ephemeral, not EBS).
How to use the seven scenarios below
Answer each one before opening the explanation, and when you miss, name the requirement that decided it — “moderate IOPS on a small volume”, “highest durability”, “sequential throughput”, “infrequently accessed”, “rebuildable”. Every explanation states that constraint and then takes the distractors apart, and the distractors are almost always a real EBS type answering the wrong bottleneck. For the domain overview, see Design High-Performing Architectures.
Updated for SAA-C03 (July 2026). The sample questions above are our own work
in the style of the exam — not real exam items. The exam itself is set and marked
by the certification body.