Connection-Level Performance

The time a handler spends building a response is only part of the time a caller waits. For example, a bookstore's handler might produce a book response in 25 milliseconds while its caller regularly waits more than 200 milliseconds. Even with a healthy database, creating a new HTTP transport for every request can add repeated connection setup costs.

Connection-level performance concerns the work a client needs to reach a service and carry requests between peers.

This chapter explains setup costs, connection reuse, pool capacity, HTTP multiplexing, and connection lifecycle management. The examples use a fictional bookstore API over HTTPS at api.bookstore.example; timing and capacity figures are illustrative.