by yoonman
WhiningDog.NET
11/15/2002

Introduction

This is an introduction to a possible series of articles that will explore various aspects of advanced windows programming using threads and I/O Completion Ports (IOCP). The article assumes you have some basic understanding of threads and winsock, however anyone with some understanding of operating systems will likely benefit. This article is based upon aspects of windows programming that I have picked up in various books (cited at the end of the article) as well as professional and academic experience. I am by far not the utmost expert in this field and in many ways I hope to be learning as much as you are by tinkering around with one of the most advanced and robust aspects of systems programming in Microsoft Windows.

The Common Paradigm

I remember one of the first server applications I made was a simple protocol server for a networking class in college. I basically had one thread waiting and listening for requests, and then upon every received request I would spawn a new thread that would handle processing and responding. Each client thread would have to wait for the I/O to return upon each communication exchange. This I/O model will work well if there are only a few clients making infrequent requests. However if you were to stress my application with hundreds of clients you would soon realize a huge degradation in performance due to the immense amount of context switching among the threads. I remember in operating systems class we learned that context switching is an expensive process since the CPU has to flush the processor pipleline before switching threads, as well as take the time to save and load registers. A much more efficient model would be to create a small amount of threads that finished jobs as fast as possible with minimal interruptions. The threads should only be switched out when the threads really did not need the processor time anymore.

[Home] [Next Page]