What System Design Interviews Test

Updated Nov 16, 2026 · 4 min read
On this page

    Two candidates are asked to design a URL shortener. Both end with almost the same diagram: an API, a database, a cache, an ID generator. One gets a strong hire; the other is rejected. The difference is not the boxes. The first asked how many links are created per day, explained why the cache helps given 100 reads per write, and spotted that a single ID counter would be a bottleneck. The second drew the same boxes in silence and answered "why Redis?" with "it's fast".

    A system design interview is scored on how you get to a design, not just on the design. Knowing what the interviewer is actually writing down changes how you spend the 50 minutes.

    How it works

    What the interviewer writes down: problem scoping, a simple high-level design that works end to end, deep dives that find the real bottleneck, trade-offs that explain why one option fits, and clear co
    What the interviewer writes down: problem scoping, a simple high-level design that works end to end, deep dives that find the real bottleneck, trade-offs that explain why one option fits, and clear communication.

    1. Problem scoping

    The prompt is deliberately vague. "Design a ticket booking system" could mean a cinema with 200 seats or a stadium concert with 500,000 fans arriving in one minute. The interviewer wants to see you narrow it down:

    • Ask about the core use cases and agree on the two or three that matter.
    • State the scale as assumptions: users, requests per second, data size.
    • Say what each number means for the design ("1,000 bookings per second on one popular event means contention on the same rows").

    2. A high-level design that works

    Within about 10–12 minutes you should have a design that handles the main use cases end to end: Client → API Gateway → Services → storage, with the main data flows. Simple and correct beats elaborate and half-finished.

    3. Deep dives

    This is where most of the signal is, especially at senior levels. You (or the interviewer) pick the parts most likely to break, such as hot keys, consistency, a slow dependency or a huge fan-out, and work through them properly: what fails, what the options are, which one fits and why.

    4. Trade-offs

    Every choice has a cost. Strong candidates say it before being asked: "Fanout on write makes reads fast but celebrity posts expensive, so I'd pull for accounts over 100,000 followers." Naming a technology is not a trade-off; explaining what you gain and give up is.

    5. Communication

    The interviewer must be able to follow you. That means thinking out loud, checking in at milestones ("does this high-level design make sense before I go deeper?"), managing time, and responding to hints instead of defending a first idea.

    Weak vs strong, side by side

    Weak versus strong answer to "the database is slow": the weak answer adds Redis and Kafka with no numbers or reasons; the strong answer states the read ratio, caches product pages with a TTL and delet
    Weak versus strong answer to "the database is slow": the weak answer adds Redis and Kafka with no numbers or reasons; the strong answer states the read ratio, caches product pages with a TTL and delete on update, and keeps stock at checkout on the database.

    The strong answer follows a pattern you can reuse: number → choice → detail → limit. The weak answer has the same technology and none of the reasoning.

    Trade-offs and costs

    The interview itself has trade-offs you manage in real time:

    • Breadth vs depth: covering every component superficially leaves no time for the deep dives that carry the most signal. Cover the basics quickly, then go deep on two or three things.
    • Leading vs following: at senior levels you should drive, but if the interviewer steers you toward a topic, follow. That is where they want the signal.
    • Precision vs speed: round numbers are fine ("about 1,000 per second"). Long arithmetic wastes time.

    When to reach for it

    Use this rubric when preparing and when practising:

    LevelWhat usually passes
    Mid-levelClear requirements, a correct high-level design, sensible choices for storage and caching, can go deeper with prompting
    SeniorAll of the above, plus leads 2–3 deep dives without prompting, quantifies, names trade-offs and failure modes
    Staff+Identifies the hardest problems unprompted, weighs alternatives against business needs, considers operations, evolution and cost

    Expectations differ between companies and teams; this table is a general guide, not any company's official rubric.

    Example

    Prompt: "Design a URL shortener." What scores well in each area:

    • Scoping: "Is it public, with custom aliases? Do links expire? I'll assume 100 million new links a month and 100 reads per write, so about 40 writes and 4,000 reads per second on average."
    • High-level design: client → API gateway → shortening service and redirect service → key-value store, with a cache in front of redirects.
    • Deep dive 1, ID generation: compare hashing the URL, a global counter, and pre-allocated ID ranges per server, and pick ranges to avoid a single counter bottleneck.
    • Deep dive 2, hot links: a link shared by a celebrity gets 50,000 reads per second; serve redirects from cache and CDN with a short TTL.
    • Trade-off stated: "301 redirects let browsers cache, which cuts load but hides click analytics; I'd use 302 if analytics matter."
    • Communication: checks in after requirements and after the high-level design, and watches the clock.

    Common mistakes

    • Treating it as a quiz with one right answer, and freezing when unsure.
    • Jumping to the architecture before asking a single question.
    • Listing technologies without saying what problem each one solves.
    • Going silent while thinking.
    • Spending the whole interview on the high-level design and never reaching a deep dive.
    • Ignoring hints from the interviewer.

    See Common Mistakes for more, with fixes.

    Say this in the interview

    "Before designing, let me confirm the core use cases and the scale. I'll state my assumptions as numbers, then get a simple design working end to end, and then we can go deep on the parts most likely to break. Stop me if you'd like to focus somewhere specific."

    Used in