INFRASTRUCTURE LEARNINGINFRASTRUCTURE LEARNING
インフラ大全インフラ大全IT INFRASTRUCTURE / LEARNING
Infrastructure learning guide
Find a lesson or lab/

Esc to close · Search terms stay on your device

LEARNING日本語

Topic lessonTopic: availability and performance

Availability vs performance: understanding redundancy and scaling

“Keep it usable” and “make it fast” need different checks. Availability concerns correct use when needed; performance concerns speed and processing volume under defined demand. Use the same report task and a fictional AWS Availability Zone (AZ) failure to distinguish passing one criterion from passing both.

What you will learn

Separate normal-performance criteria from continuity during a stated failure, judge fixed results as pass, fail or not checked, and identify further checks for redundancy and capacity changes.

Prerequisite

Understand Regions, AZs and shared versus separate failure boundaries. No EC2 or database configuration knowledge is needed. Timing runs from the client sending a request to receiving the correct response.

Separate staying usable from processing work fast enough

Availability concerns whether a workload can perform its agreed function correctly when needed. Performance concerns response speed and the amount of work processed under defined demand and operating conditions. Response time is the time from sending a request to receiving its response; throughput is work completed per unit of time. Instance counts and CPU utilization are not substitutes for the result a user needs.

To express availability as a percentage, first define an observation period, such as a month, and what counts as available; then use a measure such as available time within that period.

Start with the same report task and define two checks
Start with the same report task and define two checks1: Read the required report — Define the continuity conditions — Usable during a failure?. 2: Read the required report — Define speed and work-volume conditions — Fast enough for the demand?. 3: Usable during a failure? — Use results for the stated failure — Check matching evidence. 4: Fast enough for the demand? — Use results for the stated demand — Check matching evidence1234Read the requiredreportUsable during afailure?Fast enough for thedemand?Check matchingevidenceStart with the same report task and define two checks1: Read the required report — Define the continuity conditions — Usable during a failure?. 2: Read the required report — Define speed and work-volume conditions — Fast enough for the demand?. 3: Usable during a failure? — Use results for the stated failure — Check matching evidence. 4: Fast enough for the demand? — Use results for the stated demand — Check matching evidence1234Read therequired reportUsable during afailure?Fast enough forthe demand?Check matchingevidence
  1. Read the required report → Usable during a failure?Define the continuity conditions
  2. Read the required report → Fast enough for the demand?Define speed and work-volume conditions
  3. Usable during a failure? → Check matching evidenceUse results for the stated failure
  4. Fast enough for the demand? → Check matching evidenceUse results for the stated demand
Read the required report
Define the correct result first.
Usable during a failure?
Specify the failure, required result and acceptable wait.
Fast enough for the demand?
Specify concurrent requests and response criteria.
Check matching evidence
Placement and instance counts do not prove success.

The lines connect requirements to checks. They are not network paths, request order or failure propagation. Separating the questions makes missing evidence easier to identify.

Fixed example: two checks for the same report task

An employee needs the required internal report. Keep input size, required data and the client-side timing boundary consistent. The following two trials are fictional, not AWS measurements. One checks normal-operation performance; the other checks continued use during a specific failure.

Conditions for this lesson only; distinguish seconds from milliseconds
QuestionTrial conditionsPass criterion for this example
Normal-operation performanceNo failure. Send 3 requests simultaneously.All 3 return the correct report within 300 ms each.
Continuity during an AZ failureAZ-A is unavailable. Send 3 requests simultaneously.All 3 return the correct report within 2 seconds (2,000 ms) each.
Read the stipulated results against their matching criteria
Case3 normal-operation response times3 AZ-A-failure response timesNormal performanceFailure continuity
A150 / 220 / 300 ms; all correctNo correct response for any request by 2 secondsPassFail
B550 / 650 / 800 ms; all correct1,000 / 1,500 / 2,000 ms; all correctFailPass
C120 / 200 / 280 ms; all correct900 / 1,200 / 1,800 ms; all correctPassPass
D120 / 200 / 280 ms; all correctNo result checkedPassNot checked

A is fast in normal operation but does not continue serving the required result under the stated failure. B meets the 2-second failure condition while missing the normal-operation 300-ms condition. C meets both conditions in this example. For D, do not invent a failure result from normal-operation speed.

These are given observations, not response times predicted from a topology. No row is calculated from placement alone. A fast HTTP 200 with the wrong report is not a correct result. C exceeds 300 ms during the failure, but the current performance criterion applies only to normal operation. If users also require 300 ms during the failure, add that condition and check it separately.

Choose redundancy and scaling from the requirements

Redundancy provides alternative components or paths so that required work can continue when an element becomes unavailable. Scaling changes capacity to match demand. You can change the size of an element or change the number of elements. Consider whether required work continues after capacity decreases as well as after it increases.

A design option and what still needs checking
OptionPotential useWhat it does not establish alone
Provide processing destinations in another AZPrepare for one AZ becoming unavailableRequest failover, required data, remaining capacity and shared dependencies
Change the number of equivalent processing elementsMatch capacity to changing concurrent demandWhether a common failure stops all elements, and whether app/storage design can use the capacity
Change one element’s capacityA candidate when CPU or memory is insufficientWhether that resource is the actual bottleneck; change downtime, limits and costs
Combine request distribution with required processing destinationsUse multiple destinationsHealth criteria, app state, data and the actual user result

In AWS, Auto Scaling can maintain or change capacity, while ELB supplies request distribution as part of a design. Adding these components does not demonstrate that users’ requirements are met. State the required outcome before moving into service configuration.

Practice matching results to requirements

Choose an option to see the result and explanation.

Which criteria does fixed case B meet?
Read the answer and reason

Answer: Failure continuity only · The task is the same, but the operating condition and deadline differ. B passes failure continuity and fails normal performance.

A returns all 3 normal-operation results within 300 ms. What follows?
Read the answer and reason

Answer: It passes normal performance; read the failure evidence separately · Check normal speed and continuity during the stated failure against their corresponding results.

What does the table establish about C?
Read the answer and reason

Answer: It passes the two stated normal and AZ-A-failure conditions · C passes these two criteria. A 300-ms failure deadline, long periods, different demand and other failures remain unchecked.

Choose an option to see the result and explanation.

D has no checked failure results. How should that be recorded?
Read the answer and reason

Answer: Record failure continuity as not checked · An unavailable result remains not checked; do not replace it with either pass or fail.

What if users require C to respond within 300 ms even when AZ-A is unavailable?
Read the answer and reason

Answer: Add the criterion and check the failure results again · Separating requirements does not make them unrelated. When users’ criteria change, check evidence for those same conditions.

An HTTP 200 arrives in 120 ms, but contains another person’s report. Does it pass?
Read the answer and reason

Answer: No: the required report is incorrect · The check concerns the result the user needs. Check correctness and timing against the stated conditions.

Move from requirements to architecture checks

  1. Write what the user must complete for the result to count as correct.
  2. Specify normal demand, input and response-time conditions.
  3. Specify the failure and the required outcome and wait during it.
  4. Connect placement, failover, data, remaining capacity and shared dependencies to the requirements.
  5. Separate pass, fail and not checked; retain the periods and demand conditions not evaluated.

Choose a related lesson for your next question

Check, then explain in your own words

  1. Separate continuity from speed requirements for the same task.
  2. Match normal and failure results to their own deadlines.
  3. Identify remaining checks for redundancy, capacity changes and distribution.
Check your understanding: Does a fast normal response establish that the app remains usable during an AZ failure?

No. Check the required result and acceptable wait under that failure condition separately.

The condition used to assess speed differs from the failure condition that must be handled. Check failover, data, remaining capacity and shared dependencies as well as placement and counts.

Sources and conditions checked

Primary references are collected here with their scope, conditions and corresponding sections. Independent expert review and live-environment validation have not been performed.