Integrating fast network communication with a user-level thread scheduler

Abstract

In operating systems, traditional methods of network communication incur high overheads and usually block the process performing the communication. This means that if a process has some communication in progress, it cannot perform further computation until the communication is complete, even though the CPU is free to be used during this time. This dissertation deals with the integration of a communication system with a user-level thread scheduler. In this way, while one user-level thread is waiting for a particular communication operation to complete, other userlevel threads can perform some computation, effectively overlapping communication with computation.