【摘要】 针对流的低延迟需求,对全局延迟限制下中间件(middlebox)部署和调度问题进行建模分析。将问题形式化成一个优化问题,拆分为部署问题和流量调度问题,通过逐轮迭代的方式逐步求出整个优化问题的解。对于两个子问题,分别提出KLeveLVoting和MaskedViterbi算法,分别用贪心的方式解决部署和调度问题。为验证算法的有效性,在4种拓扑中进行算法仿真,实现传统方案的两种配置方式:静态配置(Fixed-Fixed)和简单的负载均衡(Fixed-LB)。实验结果表明,Quokka可以降低30%-50%的资源消耗并减少平均20%的延迟。更多还原

【Abstract】 Aiming at the low delay requirement of convection,the deployment and scheduling of middleware under global delay constraints were modeled and analyzed.The problem was formalized into an optimization problem,which was divided into deployment problem and traffic scheduling problem,and the solution of the whole optimization problem was gradually obtained by iteration.For the two sub-problems,KLeveLVoting and MaskedViterbi algorithms were proposed respectively,which solved the deployment and scheduling problems in a greedy way.To verify the effectiveness of the algorithm,the algorithm simulation was carried out in four topologies,and two configuration modes of the traditional scheme were realized:Fixed-Fixed and simple load balancing(Fixed-LB).Experimental results show that Quokka can reduce the resource consumption by 30%-50% as well as the delay by 20% on average.更多还原