【摘要】 本文提出的分级指导随机提早丢弃CGRED是在因特网上为用户提供有区别服务的单队列管理机制,它同时满足了对服务质量的精细控制和实现简单性两方面的需求.CGRED采用在线测量的方法跟踪业务类的实际带宽占用情况,根据给定的优先级关系确定调整业务类的指导策略,并以采用自适应算法计算得到的丢弃指导概率作为分组丢弃概率的边界,成功地解决了将不同业务类的实际带宽占用情况转化成如何丢弃不同业务类分组的实际操作.仿真结果表明CGRED是一种能有效提供DiffSery的实用网络控制机制.更多还原

【Abstract】 Class Guided Random Early Discarding (CGRED) ,a novel scheme to provide services differentiation in today’s Internet , is proposed. It can precisely maintain the assigned priority and pre-allocated bandwidth for each service class. As a single queue management mechanism, CGRED possesses all advantages of the simplicity, robustness and scalability of schemes like RIO and WRED. And it has the capability to maintain the pre-allocated bandwidth for each class of service precisely just as CBQ.The key enforcement of CGRED to standard RED is to discard packets of different service classes adaptively with different probabilities according to the bandwidth actually occupied by different classes and the pre-allocated bandwidth.To accomplish this,a couple of class-specific guidelines, the ’adjustment direction’and ’guiding probability’ ,are introduced to conduct the adaptive determination of class-differential discarding probabilities for each class of service. The guidelines for each class of service are updated periodically based on online measured bandwidth utilization. When discarding packets, CGRED sets the class-specific guidelines as the boundary of the packet dropping probability of standard RED. In this way, the unbalance of bandwidth consumption among different service classes is successfully transferred to differential operations in the buffer management among different service classes. Extensive simulation work has been conducted to verify the performance of CGRED and robustness of parameters in its algorithm. The simulation results show that CGRED is an effective scheme to provide Differential Services in networks with pre-allocated bandwidth and pre-assigned priority hierarchy among service classes.更多还原