摘要
网络流量分析表明很多信息源的聚合会产生具有自相似特性的信息流,在自相似环境下分析IP网络的排队性能成为当前热点。文中采用具有Pareto分布的ON/OFF叠加模型作为输入业务,分析IP交换机缓冲区队列溢出概率,得到与实际网络相似的结论:溢出概率并不随缓存长度的增加而呈指数规律下降,其下降速度相对要慢得多。在自相似业务环境下,增加缓存长度并不能有效地降低丢失率。
Traffic analysis of networks has shown that the aggregation of many traffic sources will produce traffic streams that are self-similar over several scales.For this reason,queuing performance study of IP networks with self-similar input traffic is a hot topic.In this paper,the input traffic is approximated by an ON/OFF aggregation model with Pareto distribution,which is an asymptotically second-order self-similar process.Compared with the typical IP traffic models currently considered in literature,the buffer overflow probability decreases non-exponentially with buffer size,but falls slowly.As a result,increasing capacity of buffer can't effectively reduce the packet loss probability.
出处
《南京邮电学院学报（自然科学版）》
2005年第3期54-58,共5页
Journal of Nanjing University of Posts and Telecommunications
关键词
自相似
重尾分布
溢出概率
Self-similarity
Heavy-tailed distribution
Overflow probability
