Exam · Throughput, latency and batching: capacity planning for a model server

Take the exam and earn your certificate.

A real, server-scored exam for this course. Score strictly above 80% to pass and get a genuine, verifiable certificate.

Before you start

We ask for your name and email first, the same as a PDF download, so we know you are here and your certificate can carry the right name.

We privately record your name, email and request IP for this exam request. Read the privacy notice and retention policy.

Visit Mickai on LinkedIn

Complete the security check before submitting.

Cloudflare Turnstile helps prevent automated submissions. How this check uses data.

You can retake the exam any time. Each attempt draws a fresh set of questions.

Written practice, ungraded

Write your response in your own notes, then compare it with the sample answer. This practice is separate from the scored exam and does not affect your certificate.

Practice 1

Rewrite the claim "the server handles 100 users" as a testable experiment question, and say how every submitted request should be accounted for.

Compare with a sample answer

A testable question is: under a specified arrival trace, prompt and output length mix and fixed engine revision, what completed request rate can the service sustain while meeting an agreed response-latency objective? Registered users are not simultaneous work, so the claim must be about requests in flight and requests per second. Every submitted request is counted as completed, failed, rejected, cancelled or unfinished, and these outcomes are reported separately. No capacity number is established until that experiment is run.

The sample replaces user counts with a workload, an objective and complete outcome accounting, without inventing a measured result.

Practice 2

Requests arrive at 0, 1 and 2 ms with capacity 2, a 3 ms batching delay and service of 8 + 2n ms. Trace the schedule and explain what a scheduler that restarts the delay when the server becomes free would get wrong.

Compare with a sample answer

The first two requests fill the batch at 1 ms and finish at 1 + 8 + 2 x 2 = 13 ms. The third request's own deadline was 2 + 3 = 5 ms, which has already passed, so it starts as soon as the server is free at 13 ms and finishes at 13 + 8 + 2 = 23 ms. A scheduler that restarted the delay at 13 ms would hold it until 16 ms and finish it at 26 ms, adding 3 ms of unnecessary waiting. The test suite includes this exact trace because a plausible-looking scheduler can make that mistake.

The sample derives each timestamp from the 8 + 2n formula and the deadline rule, and quantifies the restart defect.

Practice 3

Four requests arrive at 0, 1, 2 and 3 ms and finish at 10, 20, 30 and 40 ms. Calculate the mean latency, nearest-rank p95, completion rate and average occupancy (in two ways), and state what these numbers do not establish.

Compare with a sample answer

The latencies are 10, 19, 28 and 37 ms, which sum to 94, so the mean is 94 / 4 = 23.5 ms. Nearest-rank p95 uses rank ceiling(4 x 95 / 100) = 4, which is 37 ms. Four completions over the 40 ms window give 1,000 x 4 / 40 = 100 requests per second. Average occupancy is 94 / 40 = 2.35, and also 100 requests per second times 0.0235 seconds = 2.35. These equalities check a finite fixture; four arrivals do not establish a sustainable rate, a confidence interval or a production tail percentile.

The sample reproduces each figure from the serial burst and keeps the conclusion within what four observations can support.

Practice 4

Compare the grouped burst and the sparse trace for a product owner without making a hardware claim, and propose the next controlled comparison.

Compare with a sample answer

In the grouped burst the first request becomes slower, from 10 to 19 simulated milliseconds, while later requests benefit from sharing one fixed cost. Sparse requests find no companion within the 3 ms allowance, so each stays alone and pays an extra 3 ms. The next step is to replay representative arrival and length distributions on the actual engine, keep output-quality requirements fixed, and compare individual latency, tail latency, errors and completed throughput. The fixture does not select a universal batch size.

The sample states who gains and who loses under the assumed formula and proposes a controlled real experiment instead of a hardware conclusion.

Practice 5

Explain how the independent tick reference differs from the event simulator, and what their agreement does and does not show.

Compare with a sample answer

The tick reference advances a clock one millisecond at a time, enqueues arrivals, waits while the server is busy and dispatches when the queue is full or the oldest waiting request reaches its allowance. It does not use the event simulator's minimum of deadline and future full-batch arrival. Both versions share the declared 8 + 2n service formula and the same result record type, so their agreement checks the scheduling logic under that assumption. It does not show that the service formula models any hardware.

The sample separates what the second implementation checks (scheduling logic) from what it cannot check (the service formula against hardware).

Practice 6

Write a short handover note for the next serving experiment that another builder can run and judge.

Compare with a sample answer

The note identifies an exact engine, model and hardware configuration, a replayable arrival and request-length trace, fixed quality checks and a small chosen set of batching delays to vary. It captures per-request timing and outcome, aggregate completed throughput, agreed latency percentiles, resource observations and unfinished work. The service owner supplies the acceptance thresholds. It states that actual serving, streaming, overload, cancellation and recovery remain untested by the simulator.

The sample fixes the system and workload, varies one batching setting, names the evidence to record and keeps untested behaviour visible.

About this exam

Is this exam graded?

Yes. Every question is scored on our server, never in your browser, and you need a score strictly above 80% to pass. The questions are a real, reviewed bank for this course, drawn fresh for each attempt.

Is it only multiple choice?

Yes. This is a multiple-choice knowledge assessment, with at least one question from each chapter in every attempt. Written exercises are separate practice and are not automatically scored. Passing does not establish practical professional competence or independently verify identity.

Is there a time limit?

Yes, 45 minutes, shown as a countdown once you start. If time runs out, whatever you have answered is submitted automatically.

Do I need to download the PDF first?

No. This course's full text is on its own page, so you can read it online and come straight here when you feel ready.

What do you do with my name and email?

We privately record your name, email, request IP, course and privacy acknowledgement for this request. Newsletter consent is separate and optional. See /privacy.html for storage and retention. A passed exam creates a server-stored named certificate. An email confirmation is not identity assurance.

Can I retake it if I fail?

Yes, straight away. Each attempt draws a fresh set of questions and shuffles the answer order, so it will not be identical to your last attempt, and an attempt can only be submitted once.

Is the certificate real?

It is a genuine, server-issued certificate with its own verification page and a unique ID, not a self-declared one. It records that you passed this course's exam on this date. It is not an accredited qualification, a professional licence or proof of employment.