一种改进的核心无状态公平带宽分配机制

CSFQ(Core Stateless Fair Queue)算法在无状态网实现了如同有状态网那样好的公平带宽分配，但它的丢包算法是针对UDP流等非响应流导出的，不适用于TCP流．本文针对TCP流的特点提出了一种改进的CSFQ算法．算法主要改进有两点：(1)将缓存队列长度变化与丢包概率关联起来，用一种类似于RED(Random Early Drop)的缓存策略解决了缓存频繁溢出导致的一些问题，如公平共享速率的收敛问题；(2)对TCP流使用了与UDP流不同的丢包策略，彻底解决TCP流与UDP流的带宽分配公平性．仿真结果显示，当TCP流与UDP流共享拥塞链路时，在带宽分配的公平性方面，改进算法较原算法有了很大的性能提高．

An Improved Core Stateless Fair Bandwidth Allocation Mechanism

Through stateless networks, CSFQ (Core Stateless Fair Queue) can achieve as well fairness in bandwidth allocation as that implemented by stateful networks. However, its dropping algorithm is derived from non-responsive flows such as UDP flows, and does not suit TCP flows. An improved CSFQ algorithm was proposed in consideration of the characters of TCP. There are two main improvements in the algorithm. Firstly, the dropping probability is related to the buffer length and a buffer policy similar to RED (Random Early Drop) is used to solve such problems as the convergence of the fair share rate, resulting from frequent buffer overflow. Secondly, a dropping policy different from that of UDP is employed to TCP. Hence, the fairness of bandwidth allocation between TCP and UDP is well achieved. The simulation results show that the improved algorithm outperforms the conventional CSFQ algorithm by a significant margin in bandwidth allocation fairness when TCP flows share the congestion link with UDP flows.