[论文笔记] Reliability Prediction for Service-Oriented Computing Environments (Internet Computing, 2006)
Time: 1.5 hours
Vincenzo Grassi, Simone Patella, "Reliability Prediction for Service-Oriented Computing Environments," IEEE Internet Computing, vol. 10, no. 3, pp. 43-49, May/June 2006, doi:10.1109/MIC.2006.64 (gs:14)
作者Grassi是罗马"Tor Vergata"大学的full professor, 以下是他几篇与Web Service Reliability相关的论文: 
Reliability Modeling and Analysis of Service-Oriented Architectures. Test and Analysis of Web Services 2007: 339-362 Reliability Prediction for Service-Oriented Computing Environments. IEEE Internet Computing 10(3): 43-49 (2006) Architecture-Based Reliability Prediction for Service-Oriented Computing. WADS 2004: 279-299
以下是论文笔记:
1. 本文讨论Web Service的reliability属性, 提出了一个原型系统RelServ支持组合服务reliability的计算.
(1) No transparency
只提供一个总的reliability值. 服务提供商负责选择成员服务, 计算组合服务的reliability. 对于用户来将, 组合服务的内部结构是黑盒.
(2) Partial transparency
服务商提供service's internal reliability, its usage profile, 负责选择成员服务, 但是需要用户自己来计算组合服务的reliability.
(3) Total transparency
服务上提供组合服务的模板, 用户需要自己选择服务并计算组合服务的reliability.
3. 本文服务组合的模型
(1) 基于probabilistic flow graph
(2) 有唯一的入口和出口(Start, End)
(3) 节点表示service(可以是atomic和composite service).
Atomic service只有绿色部分, IntFailProb表示该服务本身执行时的出错率;
Compsite service还有褐色部分, 表示调用外部服务(支持AND, OR两种模式) (这里的OR是指容错模式, 而不是指condition模式). 
另: 对于composite service, reliability计算公式如下(假设服务Sk之间互相独立) 
(这两个公式比较容易理解)
4. 计算组合服务的reliability
通过将probabilistic flow graph转化为对应的failure aware flow graph(对于每一个service, 添加一个failure节点)(这个图描述了一个discrete time Markov process过程),
作者使用Markov process theory的方法来计算出End node的absorption probability(即整个组合服务的reliability).
5. 作者认为:
"Most of the approaches to the reliability prediction of component-based systems can be applied to service-oriented computing(SOC) applications with few or no modifications."
6. 感觉本文的建模方法有点奇怪, 与常见的建模方法有差异. 而且不完备.
此外在作者的一篇"Architecture-Based Reliability Prediction for Service-Oriented Computing"中, 考虑了不同的requests之间共享同一个service的情况, 从而这些requests不是互相独立的.
