摘要
基于带前缀缓存的流调度方案OBP(optimizedbatchpatching)提出了媒体后缀的增量缓存算法ICBP和预分配缓存算法PABP,推导出两种算法下传输单一流媒体对象各自所需的平均骨干带宽、服务器平均使用输出通道数和代理所需最大缓存容量,并分析了以上两种缓存算法的综合传输代价.结果表明,两种缓存算法均有效降低了骨干带宽消耗和服务器负载,传输代价均明显低于带前缀缓存的OBP.而PABP克服了ICBP算法中每次批处理调度时都至少需请求一个批处理长度补丁的缺点,因而具有更低的传输代价,其性能优于ICBP缓存算法.
Built on recently presented optimized batch patching （OBP）, two dynamic caching strategies for media suffix named IC-BP and PA-BP are proposed in this paper to alleviate the over-consumption of backbone bandwidth and tee server load in media streaming systems. Here derived are the required average backbone bandwidth, the average server channels used and the maximum cache capacity of the respective scheme when dealing with a unique media object. By defining a simple but practical cost function, the aggregate delivery cost of the two schemes that tradeoff the network and server resource are discussed. The results show that both schemes can greatly reduce the backbone bandwidth consumption and the server load, while PA-BP outperforms IC-BP with much lower cost since it saved more patch traffic by using an additional buffer to pre-buffer the incoming data in case requests arrive in the current batch interval and therefore more transmission cost is saved than IC-BP.
出处
《计算机研究与发展》
EI
CSCD
北大核心
2005年第8期1384-1390,共7页
Journal of Computer Research and Development
基金
国家自然科学基金项目(60103005)
关键词
流媒体
应用层调度
批处理
补丁
后缀缓存
streaming media
application-level scheduling
batch
patching
suffix caching
