漏桶模型下GPS调度系统的性能分析

着重分析了在数据流分别由线性漏桶和分形漏桶模型进行约束和整形的情况下，GPS(Generalized Processor Sharing)通用处理器共享调度系统的排队性能，给出了最大队长和最大时延的估算方法，并将理论分析与实验结果做了一定对比．文章分析指出，对UDP流和聚集程度不高的TCP流，用线性漏桶模型是有效的；而对于聚集程度较高的TCP流，用分形漏桶模型来描述其流量则比较合适．

On the Performance of GPS Scheduler Equipped with Different Leaky Buckets

The method for evaluating the backlog and delay bounds under GPS (Generalized Processor Sharing) scheduler is examined in detail firstly, in which linear Leaky Bucket is deployed to constrain the incoming traffic. Then an implementation based on Packet-by-Packet GPS is utilized to compare the results from analysis and measurement respectively. After that, the queueing performance of GPS scheduler equipped with Fractal Leaky Bucket, a new traffic policing mechanism which constrains traffic to the Fractional Brownian Motion envelope process is introduced briefly. Furthermore, some specific traffic traces are analyzed. The results show that linear Leaky Buckets are effective in constraining UDP-based traffic or low-aggregated TCP traffic, while the Fractal Leaky Buckets are more competent for policing the incoming traffic when the traffic comes from highly aggregated TCP flows.