Research And Implementation Of Queue Management SchemesBased On Network Processor

With the high-speed development of Internet, network link rate are increasing contiguously, while network applications and services are raising a diversity trend. Thus, the network equipments (routers, etc.) have to meet the increasing requirements of services and performance, including support for differentiated services, and flexible service extension capability, etc. Queue management is one of the key technologies for the network equipments, and network processor is also becoming the focus for the implementation platform of network equipments. So, it is of significant importance to study the design and implementation of queue management based on network processor platform. Also, it has to face several challenges and critical issues.Starting with introduction of service, control and architecture of network QoS (Quality of Service), this paper presents a survey of the research production for queue management in recent years, including summarization of the buffer management policy and packet discarding policy of queue management. At the same time, this paper also introduces several classical queue management schemes.And then, this article presents a effective and simple queue management scheme – DPBS (Dynamic Partial Buffer Sharing), and gives detailed description of the mechanisms and performance characteristic. DPBS scheme deploys partial buffer sharing policy, and adjust queue threshold based on packet loss history, which makes it achieve proportional loss rate control and rather high buffer utilization. And the most important, DPBS scheme is very easy to implement with rather low implementation overhead. For comparison, another queue management scheme – RR-PLR (Round Robin based Proportional Loss Rate) was described in this article, which can also guarantee the proportional loss rate. With Pushout policy, RR-PLR scheme can achieve highest buffer utilization; however it has rather high implementation overhead.On the IXP2400 network processor platform, we have designed and implemented a flexible and extensible queue management framework, which is used as foundation for the implementation of DPBS and RR-PLR schemes. In the implementation part of this article, we present detailed design and implementation description of the framework and queue management schemes, as well as relative issues. From the practical testing result, we can see that DPBS scheme performs better than RR-PLR scheme for its rather low implementation complexity.