# Horizontal Scaling and Backend Capacity

*MuleSoft Platform Architect Study Guide · Meeting API quality goals · Last updated November 13, 2026 · https://www.certifyforce.com/study-guide/mulesoft-platform-architect/meeting-api-quality-goals/horizontal-scaling-and-backend-capacity*

Locate client, API and backend bottlenecks before promising scale-out gains; budget replica connection pools, externalize worker-dependent state and validate representative performance tests.

- Certification: Salesforce MuleSoft Platform Architect
- Exam section: Meeting API quality goals
- Study guide: https://www.certifyforce.com/study-guide/mulesoft-platform-architect

**Official exam objective:**

> Describe when horizontal scaling of an API implementation is or is not likely to benefit response time and throughput as seen by API clients.

## What the exam expects

- Distinguish isolated response time from throughput and queueing under load.
- Locate the limiting resource in the consumer, API implementation or dependency.
- Budget outbound connections across independently configured replicas.
- Make workflows independent of a particular CloudHub worker.
- Match autoscaling conditions to measured signals and validate capacity tests.

**Horizontal scaling** adds implementation instances; **throughput** measures completed work per time; **response time** includes waiting and execution; a **connection pool** maintains reusable connections with a configured limit. An API replica and its dependency have separate resource budgets ([Backend response time](https://docs.mulesoft.com/mule-runtime/4.6/tuning-backend-server), [CloudHub availability](https://docs.mulesoft.com/cloudhub/cloudhub-fabric), [Pooling profiles](https://docs.mulesoft.com/mule-runtime/4.6/tuning-pooling-profiles)). CloudHub examples below concern classic CloudHub workers, not an assumed CloudHub 2.0 replica contract.

## Key facts

| Fact | Detail | Source |
| --- | --- | --- |
| Serial dependency | More API replicas do not divide one unchanged backend calculation. | [Backend response time](https://docs.mulesoft.com/mule-runtime/4.6/tuning-backend-server) |
| Independent requests | More workers can reduce API-side queueing without splitting a single flow. | [CloudHub availability](https://docs.mulesoft.com/cloudhub/cloudhub-fabric) |
| Backend capacity | Additional callers can increase pressure without raising dependency throughput. | [AWS retry guidance](https://aws.amazon.com/builders-library/timeouts-retries-and-backoff-with-jitter/) |
| Consumer limit | Provider replicas do not resize the client's outbound connection configuration. | [HTTP 1.9 reference](https://github.com/mulesoft/docs-connectors/blob/33276ccdb80668dd8bd45102a529a5fcc0918eea/http/1.9/modules/ROOT/pages/http-documentation.adoc) |
| Independent pools | Aggregate configured potential depends on replica count and per-pool caps. | [Database reference](https://docs.mulesoft.com/db-connector/latest/database-documentation) |
| Worker-independent state | External files and workflow references avoid dependence on one worker. | [CloudHub development](https://docs.mulesoft.com/cloudhub/developing-applications-for-cloudhub) |
| Scaling signal | A higher maximum is a range change, not satisfaction of a CPU condition. | [CloudHub autoscaling](https://docs.mulesoft.com/cloudhub/autoscaling-in-cloudhub) |
| Tool consistency | Changing the test tool requires executing all tests again. | [Performance validation](https://docs.mulesoft.com/mule-runtime/4.6/tuning-test-validations) |

## Check yourself

- Why verify functional behavior before performance testing?
- When the load-test tool changes, which tests must run again?
- Why review ramp-up, tear-down and error-percentage settings before test execution?

## Sources

- [Backend Server Response Time — Mule Runtime 4.6](https://docs.mulesoft.com/mule-runtime/4.6/tuning-backend-server)
- [CloudHub High Availability Features](https://docs.mulesoft.com/cloudhub/cloudhub-fabric)
- [Timeouts, retries, and backoff with jitter](https://aws.amazon.com/builders-library/timeouts-retries-and-backoff-with-jitter/)
- [HTTP Connector 1.9 Reference, February 2024 source](https://github.com/mulesoft/docs-connectors/blob/33276ccdb80668dd8bd45102a529a5fcc0918eea/http/1.9/modules/ROOT/pages/http-documentation.adoc)
- [Pooling Profiles — Mule Runtime 4.6](https://docs.mulesoft.com/mule-runtime/4.6/tuning-pooling-profiles)
- [Database Connector Reference](https://docs.mulesoft.com/db-connector/latest/database-documentation)
- [Develop Applications for CloudHub](https://docs.mulesoft.com/cloudhub/developing-applications-for-cloudhub)
- [Autoscaling in CloudHub](https://docs.mulesoft.com/cloudhub/autoscaling-in-cloudhub)
- [Performance Test Validations — Mule Runtime 4.6](https://docs.mulesoft.com/mule-runtime/4.6/tuning-test-validations)

## About the full unit

The "Core concepts", "Exam traps" and "Worked scenario" sections of this unit, its 8 linked practice questions, and unlimited timed mock exams are included with the MuleSoft Platform Architect pack (USD $19): https://www.certifyforce.com/practice-exams