A good API design is important in a system architecture because all data exchange between services happens either through messages or API calls. APIs must be efficient to avoid creating chatty I/O. Because services are designed by teams working independently, APIs must have well-defined semantics and versioning schemes, so that updates don’t break other services.

It’s important to distinguish between two types of API:

• Public APIs that client applications call.
• Backend APIs that are used for interservice communication.

These two use cases have somewhat different requirements. A public API must be compatible with client applications, typically browser applications or native mobile applications. Most of the time that means the public API will use REST over HTTP. For the backend APIs, however, you need to take network performance into account. Depending on the granularity of your services, inter-service communication can result in a lot of network traffic. Services can quickly become I/O bound. For that reason, considerations such as serialization speed and payload size become more important. Some popular alternatives to using REST over HTTP include RPC, Apache Avro, and Apache Thrift. These protocols support binary serialization and are generally more efficient than HTTP.

Source: Designing APIs for microservices