Improving reactivity to I/O events in multithreaded environments using a uniform, scheduler-centric API

Abstract

Reactivity to I/O events is a crucial factor for the performance of modern multithreaded distributed systems. In our schedulercentric approach, an application detects I/O events by requesting a service from a detection server, through a simple, uniform API. We show that a good choice for this detection server is the thread scheduler. This approach simplifies application programming, significantly improves performance, and provides a much tighter control on reactivity.