軟體發展流程之最佳化

Abstract

近年來，國內外軟體產業逐漸意識到軟體開發成功關鍵不僅在於是否使用新技術，更在於能否妥善管理軟體開發流程。為了提升軟體品質及落實軟體工程，目前各國軟體產業正大力推廣能力成熟度整合模式(Capability Maturity Model Integrated, CMMI)。CMMI評鑑制度共分為五個等級；目前各國已有多家軟體廠商通過初級評鑑，但在導入高等級成熟度時面臨許多難題。為了協助軟體廠商通過較高等級成熟度之評鑑，國內外文獻提出許多方法，包括統計製程管制(Statistical Process Control, SPC)和預測模型等目前常用的方法以及應用一些量化模擬方法來解決軟體發展流程最佳化之問題。然而，在軟體發展流程最佳化方面，現有的量化方法(如：隨機最佳化方法)模擬理論複雜、相關軟體不易學習且演算過程非常耗時。因此，本研究利用變異數分析法發展一套找尋各子流程的最佳方案之演算法，不僅能達成現有文獻建議的隨機最佳化方法之功用，且遠較模擬方法容易操作與學習。針對軟體發展流程之多個績效指標同時最佳化問題，目前尚無相關文獻，本研究利用多變量變異數分析法發展出一套可同時最佳化多個績效指標的演算方法。本論文最後以文獻上一個軟體發展流程之實際資料為例，來說明本研究方法確實有效可行。

In recent years, the key reasons for the success in software industry are the capability of managing software development process in addition to the use of new technology. In order to improve the software development process (SDP), Capability Maturity Model Integration (CMMI) is widely adopted by the software companies all over the world. The staged approach of CMMI yields appraisal results as one of five maturity levels. Many organizations have conducted the appraisal of maturity level 2 or 3. However, when software organizations continuously enhance their software process maturity to maturity level 4 or 5, most of them are confronted with some difficulties. Many studies proposed methods, including Statistical Process Control (SPC), prediction models and some quantitative simulation methods, to help software organizations to meet the requirements of maturity level 4 or 5. However, some of these quantitative methods, such as stochastic optimization modeling (SOM), are too complicated for software engineers or managers to learn and employ. Another drawback is that adopting these methods is too time-consuming. To simplify the process of optimizing the single performance-indicator, this study utilizes the analysis of variance (ANOVA) to determine the best option of each sub-process. The proposed method can not only achieve the effect as SOM but also save a lot of time or effort as compared with SOM. In addition, this study utilizes the multivariate analysis of variance (MANOVA) to optimize multi-performance-indicator simultaneously and determine the best option of each sub-process. Finally, a real case from a paper of SEI is utilized to demonstrate the effectiveness of the proposed procedure.