Skip to content

Sizing a thread pool for work that waits

A service on 4 cores handles requests that spend about 90% of their time waiting on a database. How should its thread pool be sized?