大规模流媒体服务间隔缓存策略的性能预测模型

静态分析模型能在负载固定的情况下预期不同因素对缓存算法性能的影响,为解决该模型在动态负载下预测精度不高的问题,结合实际用户访问行为研究,采用负载拟合的方法对此问题进行探讨,并提出了一种针对变动负载的间隔缓存类算法的动态性能模型。该模型可更准确估算系统实际性能,从而为用户控制等策略提供参考。实验结果表明,静态模型的缓存命中率预测结果比实际高70%以上,而该模型则能适应负载的变动,预测结果与实际结果差别在10%左右。

Dynamic cache performance analytical model for large video services

Caching is critical in video streaming services with interval caching as an effective caching policy.A static analytical model has been developed to estimate the performance of interval caching and to evaluate the impact of various parameters on the performance of interval caching schemes assuming a stable stream arrival rate.However,in commercial video service systems,the arrival rates of the various streams varies from hour to hour.This paper presents a dynamic model for real loadings for interval-based policies which can be used in dynamic loading environments.The model has been extensively validated for a range of workloads and other parameters and can be used to estimate the performance of real systems to get better access control.The performance of the static model differs from the actual results by 70% while the performance of the dynamic model differs by only 10%.