Research And Implementation Of The Adaptive Network I/O Model

With the continuous development of the rapid innovation in IT, more and morecompaniesdevelop a good network products include a variety of software applications orlargegame, there are many physical communication equipment. These products haveacontinuity that is a client-server model, a large number of clients communicate with server,to complete the entire product’s features. More and more Internet companies not only selltheir own products, more products are sold as client, and then companies can keep in touchwith them using servers. So that companies not only just sell a service of products, but alsocan get the information of users, which is the most important thing they care about. Once theyget the information, they can provide maintenance and upgrade work conveniently, also canuse the data and information to do data mining, and foundpotential business opportunities.It’s not difficult to see the importance of network communication between server andclient.socket is used to communicate beteen server and client, different platforms havedifferent implementations. this research is a network communication on the unix/linuxplatform.There’re five kinds of network I/O Model, normally: blocking、non-blocking I/Omultiplexing、signal-driven I/O and asynchronous I/O. Each model has its most suitableoccasion, has alot of difference in the performance.So it’s important to select a suitable andefficient model for a client-server model application, because it affects the performance ofyour application.This topic firt introduces the principle of these types of I/O model, as well as theirdifferences,and highlight select model and epoll model,discript implementation、advantagesand disadvantages between them. Then introduce a telecommunications project. In the processof development and testing of the actual project, which led me to find the idea of adaptivenetwork I/O model, this adaptive model is available on seelct model and epoll model,andshow better performance than both of them.By the next step, study the mature network communications code. On the basis ofunderstanding and learning, we found they have advantages in cross-platform、high-concurrency、high-performance and high stability, but they are only selected certain kindof network I/O model at the bottoma, and once selected the model, the running process can not be changed.This paper mainly introduce two commonly used network I/O model: select、epoll. On thebasis of comparing their advantages and disadvantages, improve each model, and put themtogether to design and realizate an adaptive network I/O model. Using amounts of data toverify that this adaptive model could has excellent performance in practical applications.Adaptive network I/O model is a new network communication model, which learns fromselect model and epoll model. Using this model, server is able to self-select the best way torunning according to the intensity of the client data, which improve the performance of theserver.