# Performance and scaling

Source: https://amplitude.com/docs/feature-experiment/under-the-hood/experiment-performance-scaling

---

On this page

- [Implementation recommendations](#implementation-recommendations)
- [Architecture](#architecture)

# Performance and scaling

To run a powerful, insight-generating experimentation program, you’ll need three things:

- A robust identity resolution system
- Access to a store of user metadata
- Access to behavioral cohorts

If you’re looking to launch an experimentation program at your company, you should know that **Amplitude Experiment has all three**.

To get the most out of these capabilities, make a request to Amplitude Experiment’s endpoint for each individual user.

## Implementation recommendations

To provide the best experience to your users, we recommend the following:

- **Use local defaults**. All Amplitude SDKs support local defaults for experiments. To accommodate the rare occurrence that Amplitude’s systems might be down, we recommend using local defaults for all your experiments
- **Use local storage (cache) on Client SDKs**. Amplitude’s client-side SDKs store user variants in local storage. This can help reduce the number of network calls coming from the client. Read more about performance and caching [here](https://www.docs.developers.amplitude.com/experiment/general/performance-and-caching/).

## Architecture

The Amplitude Experiment architecture is built on two components:

- **Fastly CDN**. All requests to Amplitude servers are routed through Fastly, one of the best CDNs out there.
- **Reliable hosted AWS services**. Amplitude uses Application Load Balancer, Relational Databases, and DynamoDB, which guarantee high availability.

Was this helpful?

<!--$-->

<!--/$-->