一利用參考機制來預測QoS的方法

Abstract

在service-based systems的開發中，QoS是一個重要的考量因素來決定要將哪些現存的服務透過組合和架構良好的流程圖，產生新的服務。由於在現實的環境中，QoS是藉由收集歷史紀錄來獲得的，通常有許多QoS是未知的。為了能更好的選擇出合適的服務，許多預測QoS的方法被提出，再藉由此預測值來做選擇的依據。然而，現存的預測方法在歷史紀錄缺乏時，其準確度會非常低，在選擇服務上無法達到原本的目的。在這篇論文中，我們提出了一個建立虛擬平台的方法，並藉由參考其資訊，來增加現有的記錄數量，以達到更好的預測結果。我們也做多項實驗來評估提出的方法，實驗數據顯示我們的預測結果在一般的情況下準確度皆優於現存的方法。

In service-based systems (SBS), the quality of services (QoS) plays an important role for application development by helping select more suitable services, based on QoS values, for those specified inside a workflow. There are several QoS prediction approaches proposed. However, the prediction accuracy is low when there are lack of historical records in the application environment. In this thesis, we propose a method to construct a virtual platform based on Gaussian distribution with stability and performance, and design a referral-based QoS prediction method associated with the virtual platform to improve prediction accuracy. The experimental results indicate that our approach is better than others with higher prediction accuracy.