虚拟延迟矢量核心无状态调度策略VCSVC

VCSVC: Vector Core Stateless Virtual Clock

提出一种新的核心无状态调度策略VCSVC(V ector Core S tate less V irtua l C lock),可实现与VC相同的延迟特性,并提供逐点精确的基本(预留)带宽保证.后者是现有的核心无状态调度算法所不能保证的,为实现核心无状态下的公平调度策略提供了可能.VCSVC中,在边界节点计算的虚拟延迟矢量作为流状态带在分组头中,是核心节点排序的依据.核心节点的处理开销主要是排序操作,具有较高的可扩展性.为了限制虚拟延迟矢量编码长度,本文提出距离粒度法,证明了由此带来的误差对延迟特性无影响,对带宽分配的精确性影响不大.

Proposed a new core stateless packet scheduling algorithm that can allocate accurate reserved bandwidth to each flow at each traversed node, called VCSVC （Vector Core Stateless Virtual Clock）. This algorithm lays the foundation for imple-menting core stateless fair queuing in deterministic ways that those existing core stateless algorithms can＇t guarantee. In VCSVC, the ingress node will compute virtual delays for each injected packet at every node along its path and encode them as a virtual delay vector into the packet header. All core nodes serve packets according to their virtual delays. Core nodes don＇t perform per flow management, and therefore have high scalahility. The distance granularity approach was presented to bound the encoding length of the virtual delay vector and the extra delay brought hy this method was analyzed.