机构地区: 清华大学信息科学技术学院计算机科学与技术系
出 处: 《清华大学学报(自然科学版)》 2006年第7期1293-1296,共4页
摘 要: 现有提供确定性性能保证的核心无状态分组调度策略中,可扩展性的获得通常以中间节点的速率损失为代价,从而导致数据传输突发性的增强,降低了网络资源的利用率。为了减少速率损失,提出一个核心无状态调度框架DVRS(de lay vector reference system),通过运用虚拟延迟矢量技术,在保证端到端延迟上限与V irtua lC lock算法相同的同时,将核心无状态算法的基本速率保证能力从现在的首尾点精确保证提高到逐点精确保证;并利用局部平均法满足用户对处理开销与带宽保证精度的不同需求;另外,由于核心节点操作的相似性,DVRS类算法具有较好的配置渐进性。 In existing deterministic core-stateless packet scheduling schemes, scalability is often achieved at the expense of under-utilization of the reserved bandwidth for flows at core nodes. This increases the burstiness in the traffic flow and limits the utility of network resources. The under-utilization is reduced by a core-stateless packet scheduling architecture, called the delay vector reference system. By using the virtual delay vector technique, the system provides the same end-to-end delay bound as VirtualClock and guarantees per-node accurate reserved rate allocation for each flow not just at edge nodes as with existing core-stateless schemes. Also, the system uses the partial average approach to meet a variety of accuracy and cost requirements. Additionally, these algorithms have good incremental deployability due to the similarity of operations at core nodes.
关 键 词: 计算机网络 分组调度 核心无状态 虚拟延迟矢量
领 域: [自动化与计算机技术] [自动化与计算机技术]