To support quality of service （QoS） management on current Internet working with best effort,we bring forth a systematic approach for end-to-end QoS diagnosis and quantitative guarantee. For QoS diagnosis,we take contexts of a service into consideration in a comprehensive way that is realized by exploiting causal relationships between a QoS metric and its contexts with the help of Bayesian network （BN） structure learning. Context discretization algorithm and node ordering algorithm are proposed to facilitate BN structure learning. The QoS metric is diagnosed to be causally related to its causal contexts,and the QoS metric can be quantitatively guaranteed by its causal contexts. For quantitative QoS guarantee,those causal relationships are first modeled quantitatively by BN parameter learning. Then,the QoS metric is guaranteed to certain value with a probability given its causal contexts tuned to suitable values,that is,quantitative QoS guarantee is reached. Simulations with three sequential stages：context discretization,QoS diagnosis and quantitative QoS guarantee,on a peer-to-peer （P2P） network,are discussed and our approach is validated to be effective.
LIN Xiang-tao,CHNEG Bo,CHEN Jun-liang,QIAO Xiu-quan State Key Laboratory of Networking and Switching Technology,Beijing University of Posts and Telecommunications,Beijing 100876,China
提出了一种基于代理缓存的移动流媒体动态调度算法DS2AM2PC(Dynamic Scheduling Algorithm for Mobile Streaming Mediabased on Proxy Caching),采用代理缓存窗口自适应伸缩和分段缓存补丁块方案,在代理缓存中根据具体情况每次缓存相同或者不同大小的段补丁块,同时隔一段时间,根据移动媒体流行度更新一次缓存窗口大小,动态决定其最大缓存大小,实现了移动流媒体对象在代理服务器中缓存的数据量和其流行度成正比的原则.仿真结果表明,对于客户请求到达速率的变化,DS2AM2PC算法比P3S2A(Proxy-assisted Patch Pre-fetching and Service Scheduling Algorithm)算法和OBP(Optimized Batch Patching)+prefix & patchcaching算法具有更好的适应性,在最大缓存空间相同的情况下,能显著减少通过补丁通道传输的补丁数据,从而降低了服务器和骨干网络带宽的使用,能快速缓存媒体对象到缓存窗口,同时减少了代理服务器的缓存平均占有量.
