CDN based on P2P and monitoring the transport subsystem design (1)

1 Introduction

P2P peer-to-peer network (Peer-to-peer) technology is currently the international computer network technology field study of a hotspot.

The technology of embryonic form in the 1970s, is a typical representative of UseNet and FidoNet, while CDN content distribution network (Content Distri-bution Network) is the site's content or media publishing to the closest to the user's network "edge", when a user visits, the system automatically seamless to the user is redirected to the edge servers, thereby reducing the hub server and backbone network, enhance streaming or Web site performance.

With the rapid development of Internet technology, streaming media content on the Internet in a large number of communication, for high-quality streaming media distribution services reflect the more obvious, therefore, was to a large number of users with fast, high-quality streaming media distribution services has become a recent study of hot and difficult.

In multimedia monitoring system in the amount of data to be transferred is fairly large, mainly including: control information, feedback, information, video, audio, and other such text information etc.

For traditional patterns based on C/S or B/S mode multimedia monitoring system, in control points and control points for these large amounts of streaming media data transmission lines, server performance will decrease. This article introduces the P2P technology to the multimedia control system design is mainly the following improvements:

(1) design a CDN based on P2P and monitoring the transport subsystem.

(2) the customer uses the P2P way receive service from the edge server, while the original server and the edge server content published via P2P manner, by means of effective use of the system of network bandwidth and host resources, thereby reducing the original server and the edge servers, reduces the backbone data flow, reducing operator costs, improve customer service quality.

(3) in order to mitigate network i/o, and disk I/O of contradictions, in the transport subsystem design use semi-synchronous/half asynchronous way network I/O and disk I/O, and through the task of the buffer pool.

(4) design a thread pool to dynamically manage algorithm, effective reduces CPU load pressure, increasing network throughput and system overall performance.

(5) against the disadvantages of traditional methods for effective improvement, use semi-synchronous/half an asynchronous way system framework, use task pool on data read or write request for packaging, using a thread pool tasks to task pool for efficient asynchronous processing.

The tasks of idle situation, combined with the system's current resource utilization, on task pool and thread pool to dynamically manage, reduces CPU load and increases system throughput.

2 system framework

System overall layout as shown in Figure 1, edge servers will be formed with a number of client nodes, P2P network to provide efficient service quality, easy to reduce the load on the server.

When a client requests an edge server, the resource does not hit the edge server to request the original server, depending on the original server, you will need to request media resource efficient by article implementation of the transport subsystem are stored locally, and then use P2P way to publish multiple edge servers.

In this way effectively reduces the original server in the content you publish.

In theory, it just adds a full media copies sent out, the other edge servers will be based on the P2P a complete copy. Similarly, if the edge servers provide the customer with service, theoretically it would only need to transmit a copy of the multiple clients can get the complete service. The original server and media resource server typically is on a subnet, network faster than disk I/O speed more quickly. At this point, the disk I/O bottleneck to the system. In order to alleviate network I/O and disk 1/O contradiction in the transport subsystem design which use semi-synchronous/half asynchronous way network I/O and disk I/O, left, and through the task of the buffer pool.

Upper main Threading epoll asynchronous events and Protocol interactions, the framework the received data according to the fixed-size wrappers to the task, and then set the task to task pool, the lower the thread pool is responsible to remove the task from a task, the pool for specific disk reads and writes, the operation is completed threads and tasks were to return to the thread pool and the pool waiting for scheduling a task.

3 algorithm

In order to thread pool for effective dynamic management, you need to capture various performance parameters, after a comprehensive analysis, make adjustments to the thread pool.

The algorithm in reference to the two most critical parameter, the task of the average wait time and CPU usage. By the average wait time for tasks, you can get the current thread pool analysis need to be adjusted. By CPU usage can be the need to increase or reduce the thread.

[1] [2] [3] [4]