應用統計理論與統計製程管制方法建構軟體產業CMMI Level-4量化管理流程績效基準及績效預測模型

Abstract

從全球資訊業發展趨勢與各項統計數據報告中顯示，中國大陸與印度的軟體產業正 蓬勃發展，究其成功的主要因素，除了該國政府長期培養優秀人才外，積極提升軟體產 業之品質，也是成功的另一個主因。在軟體產業流程管理中代表軟體品質與開發管理能 力的能力成熟度整合模型(Capability Maturity Model Integration, CMMI)認證，是各大 軟體廠商專案承接與執行能力的國際級證明。我國政府有鑑於此，大力推動「提升資訊 軟體品質(CMMI)計畫」，其目標係希望透此計畫於2008年促成70家廠商達到CMMI Level-3等級與3家廠商達到CMMI Level-5等級。然而檢視2007年11月的統計數據，發現 國內通過CMMI評鑑的組織單位，Level-2等級有49家，Level-3僅有19家，對於達成CMMI 計畫目標仍有一段差距，尤其在CMMI Level-4量化管理級方面，台灣目前尚未有任何一 組織單位通過評鑑。其問題在於無法正確的使用統計理論與方法達到CMMI Level-3與 Level-4量化專案管理的目標，究其最主要原因是這些軟體產業之組織單位在應用統計製 程管制(SPC)方法時忽略了軟體產業與製造業有基本上的差異，軟體發展是個過程，每 個過程的產出都牽涉到人而非機器，換言之，軟體發展是人員高度密集的生產過程，因 此必須充分了解軟體發展之各個階段的背景資訊、所重視之度量(metric)與組織目標，才 能有效且正確地將統計方法應用於軟體產業之流程管理中。由於中外文獻關於這方面之 研究並不多見，但上述議題卻是軟體產業提升品質的重要研究，因此本計畫擬透過統計 理論與方法，從軟體產業CMMI的量化專案管理的角度建立一套完整的流程與制度，供 資訊軟體產業改善軟體流程，以提升其品質並強化成熟度等級。本計畫共分三年完成， 第一年的工作是從探討軟體產業與製造業的異同著手，調查國內軟體產業在量化管理上 的困難，深入了解在軟體專案管理中使用統計理論與統計製程管制與這些方法應用在製 造業時不同之處。本計畫第二年的工作是，發展及檢討軟體發展中各個階段的績效度量 指標，利用各種SPC手法建立各重要指標之流程績效基準(performance baseline)，以作為 組織量化管理的基準，並應用此基準監控未來專案與流程的進行。本計畫第三年的工作 是，利用發展之流程績效指標，透過穩定的歷史資料來建立軟體流程中績效指標之預測 模型，此模型可預測在各個階段所關注的軟體發展績效值。最後，將本計畫成果撰寫成 軟體程式，以利軟體產業組織或企業能快速提升其品質至CMMI Level-4成熟度。本計畫 所使用之數據乃根據企業組織所提供的實務資料以及模擬數據，同時考量理論與實務上 的限制，因此透過本計畫研究成果及所提出流程、步驟與模型，可作為企業組織流程量 化管理之重要參考依據，在提升國內軟體產業之軟體品質、流程改善與組織成熟度上， 兼具學術及應用價值。

The development trends of global information industry and related statistics reports indicate that software industry of Mainland China and India is booming. The key success factors lie not only in cultivating professionals by their government but also in enhancing software quality. The large software companies in these two countries demonstrate the outstanding process performance and abilities to undertake transnational projects by applying Capability Maturity Model Integration (CMMI). CMMI is a process improvement maturity model for the development of products or services commonly used in system engineering and software engineering. Many organizations find value in measuring their progress by conducting an appraisal and earning a maturity level rating or a capability level achievement profile. Seeing that our government realized the focus of global competitions has been transformed from the competition of quantity and physical stamina to one of innovation and quality. Therefore, since 2005, the government has started to promote software-quality elevation project, with CMMI at its core. The main objectives of this project in the period between July 1st, 2005 and Dec. 31st, 2008 are to achieve 70 business enterprises certified as CMMI level 3, and 3 enterprises as CMMI level 5. However, according to Industrial Development Bureau data till the end of Nov. 2007, 49 Taiwanese software companies have certified as CMMI level 3 and only 19 companies certified as CMMI level 4. Besides, no Taiwanese organizations or enterprises have certified as CMMI level 4. Consequently, looking into the main reason of the phenomenon is inability to manage projects quantitatively through statistical methods in most of software companies. The major differences between software industry and manufacturing industry are software product produced by individuals not machines, software process has significant sources of variability that can be difficult to minimize or control during development and software development activities are generally more process-based than product-based, making it difficult to apply statistical process control (SPC) or other statistical methods in a straight forward manner. Moreover, the inability to select appropriate metrics in software processes will reduce successful implementation of the SPC method in organizations, ultimately inhibiting the maturity of processes management. Based on the above, this study will develop procedures and mechanisms based on statistical theory and methods to assist software companies to enhance software quality and achieve the goal of quantitatively managed. This three-year study is divided into three phases. In the first year, literatures are reviewed and a survey for software companies is carried out to investigate the core problems of quantitatively managed process in software industry. Practical data and simulated data are collected and generated as well in this phase. In the second year, several SPC methods are used to construct process performance baselines in order to be the benchmarks for the quantitatively managed process. The baselines can be applied to monitor project progress and process performance in the future. In the third year, statistical prediction approaches (e.g. stepwise regression, time series, artificial neural network and reliability method) based on stable historical data are used to construct process performance prediction model. Moreover, real cases from software industry are utilized to demonstrate the effectiveness of the proposed method. Finally, this study can assist Taiwanese organizations and enterprises in maturing software processes and enhancing software quality.