---
topic: system-design
author: Crashtech Editorial
published: Nov 13, 2026 · read: 3 min
---

Eight Non-Functional Requirements, Each Paired With the Mechanism That Delivers It

Availability, latency, scalability, durability, consistency, modularity, configurability, resiliency: how to state each measurably and what provides it.

– –

Requirements meetings are full of “it should be fast and reliable”. That is not a requirement, it is a mood. The useful move is to pair each quality with a measurable target and the mechanism that earns it, then notice what each one costs.

The eight, with mechanism and a measurable form

QualityPlain meaningTypical mechanismExample target
AvailabilityUsable when neededRedundant instances behind load balancers, health checks, multi-zone99.9% monthly
LatencyTime to respondCDNs, caching, regional placement, efficient queriesp95 under 300 ms
ScalabilityHandles growthStateless services, replication, partitioning10x traffic with flat latency
DurabilityCommitted data survivesTransaction logs (write-ahead), replication, backupsNo loss of committed writes
ConsistencyEveryone sees the same dataChosen model: strong or eventual consistencyRead-your-writes on profile edits
ModularityParts change independentlyLoose coupling, high cohesion, clear interfacesA change touches one module
ConfigurabilityBehaviour changes without codeConfiguration as code, feature flagsConfig change deploys without a release
ResiliencyRecovers from failureMessage queues, retries, circuit breakers, failoverRecovery within 5 minutes

How each mechanism actually works

Availability with load balancers remove single points of failure
Traffic spreads across several instances, and unhealthy ones are removed from rotation. The load balancer itself must be redundant, or you have moved the single point of failure.
Latency with CDNs and caches move data closer
A CDN serves content from nearby edge locations. Caches keep hot data in fast memory. Both trade freshness for speed.
Scalability with replication more nodes, more throughput
Copies of data across nodes let reads spread out. Writes still need partitioning once one primary cannot keep up.
Durability with transaction logs write it down first
Changes are appended to a log before they are applied, so after a crash the system replays the log and reconstructs committed state.
Consistency via eventual consistency accept temporary differences
Replicas may disagree briefly and converge over time. You gain availability and latency and give up instant agreement.
Resiliency with message queues buffer and retry
A queue decouples producer from consumer, absorbs spikes and lets failed work be retried instead of lost.

Modularity and configurability are the two that live in code and process rather than infrastructure. Modularity comes from loose coupling (modules know little about each other) and high cohesion (each does one job). Configurability comes from keeping settings in version-controlled configuration as code, so behaviour changes without editing core logic.

Every one of these costs something

Redundancy costs money. Caching costs freshness. Eventual consistency costs simplicity of reasoning. Queues cost visibility into when work finishes. Naming the cost next to the mechanism is what turns a diagram into a decision.

Advertisement

Rank them before you design

Qualities conflict, so choose an order for your product.

SystemRises to the topAccepted trade
Payments ledgerDurability, consistencyHigher latency, lower write throughput
Social feedAvailability, latencyStale reads, eventual consistency
Internal admin toolModularity, configurabilityModest scale, simple infrastructure
Video streamingAvailability, scalabilityComplex caching and edge layers

Do

Write each requirement as a number with a window, check it in production with an SLO, and review the ranking when the product changes.

Don't

Ask for “five nines, instant consistency and low cost” in one breath. Those goals contradict, and the design will quietly sacrifice one without telling you which.

Takeaway

Non-functional requirements are where architecture is actually decided. Put a number on each, pick the mechanism that provides it, write down what it costs, and rank them for your product. That is a design conversation. “Make it scalable” is not.

Advertisement

Frequently asked questions

What is a non-functional requirement?

A quality attribute describing how well the system must behave rather than what it does: how available, fast, durable, consistent or maintainable it is. Functional requirements say a user can place an order. Non-functional requirements say it works at 99.9 percent availability within 300 milliseconds.

How do I make a non-functional requirement measurable?

State a number and a window: availability 99.9 percent per month, p95 latency under 300 milliseconds for checkout, recovery point of 5 minutes. A vague word like fast or reliable cannot be tested, monitored or traded off. Service level objectives are the usual way to write them down.

Do all eight qualities matter equally for every system?

No, and they often conflict. Strong consistency costs latency, extreme availability costs money, high configurability costs simplicity. A payments system prioritizes durability and consistency, a social feed prioritizes availability and latency. Rank them for your product before choosing mechanisms.

What is the difference between availability and resiliency?

Availability is the fraction of time the system is usable. Resiliency is its ability to absorb failures and recover: retries, queues, circuit breakers, failover. Resiliency is one way to reach availability, but you can be available simply by being lucky, and resilient yet briefly unavailable during a failover.

Sources & further reading

/* Comments */