Software development in the post-PC era : towards software development as a service

Abstract

Engineering software systems is a complex task which involves various stakeholders and requires planning and management to succeed. As the role of software in our daily life is increasing, the complexity of software systems is increasing. Throughout the short history of software engineering as a discipline, the development practises and methods have rapidly evolved to seize opportunities enabled by new technologies (e.g., the Internet) and to overcome economical challenges (e.g., the need for cheaper and faster development). Today, we are witnessing the Post-PC era. An era which is characterised by mobility and services. An era which removes organisational and geographical boundaries. An era which changes the functionality of software systems and requires alternative methods for conceiving them. In this thesis, we envision to execute software development processes in the cloud. Software processes have a software production aspect and a management aspect. To the best of our knowledge, there are no academic nor industrial solutions supporting the entire software development process life-cycle(from both production and management aspects and its tool-chain execution in the cloud. Our vision is to use the cloud economies of scale and leverage Model-Driven Engineering (MDE) to integrate production and management aspects into the development process. Since software processes are seen as workflows, we investigate using existing Workflow Management Systems to execute software processes and we find that these systems are not suitable. Therefore, we propose a reference architecture for Software Development as a Service (SDaaS). The SDaaS reference architecture is the first proposal which fully supports development of complex software systems in the cloud. In addition to the reference architecture, we investigate three specific related challenges and propose novel solutions addressing them. These challenges are: Modelling & enacting cloud-based executable software processes. Executing software processes in the cloud can bring several benefits to software develop ment. In this thesis, we discuss the benefits and considerations of cloud-based software processes and introduce a modelling language for modelling such processes. We refer to this language as EXE-SPEM. It extends the Software and Systems Process Engineering (SPEM2.0) OMG standard to support creating cloudbased executable software process models. Since EXE-SPEM is a visual modelling language, we introduce an XML notation to represent EXE-SPEM models in a machine-readable format and provide mapping rules from EXE-SPEM to this notation. We demonstrate this approach by modelling an example software process using EXE-SPEM and mapping it to the XML notation. Software process models expressed in this XML format can then be enacted in the proposed SDaaS architecture. Cost-e cient scheduling of software processes execution in the cloud. Software process models are enacted in the SDaaS architecture as workflows. We refer to them sometimes as Software Workflows. Once we have executable software process models, we need to schedule them for execution. In a setting where multiple software workflows (and their activities) compete for shared computational resources (workflow engines), scheduling workflow execution becomes important. Workflow scheduling is an NP-hard problem which refers to the allocation of su cient resources (human or computational) to workflow activities. The schedule impacts the workflow makespan (execution time) and cost as well as the computational resources utilisation. The target of the scheduling is to reduce the process execution cost in the cloud without significantly a ecting the process makespan while satisfying the special requirements of each process activity (e.g., executing on a private cloud). We adapt three workflow scheduling algorithms to fit for SDaaS and propose a fourth one; the Proportional Adaptive Task Schedule. The algorithms are then evaluated through simulation. The simulation results show that the our proposed algorithm saves between 19.74% and 45.78% of the execution cost, provides best resource (VM) utilisation and provides the second best makespan compared to the other presented algorithms. Evaluating the SDaaS architecture using a case study from the safety-critical systems domain. To evaluate the proposed SDaaS reference architecture, we instantiate a proof-of-concept implementation of the architecture. This imple mentation is then used to enact safety-critical processes as a case study. Engineering safety-critical systems is a complex task which involves multiple stakeholders. It requires shared and scalable computation to systematically involve geographically distributed teams. In this case study, we use EXE-SPEM to model a portion of a process (namely; the Preliminary System Safety Assessment - PSSA) adapted from the ARP4761 [2] aerospace standard. Then, we enact this process model in the proof-of-concept SDaaS implementation. By using the SDaaS architecture, we demonstrate the feasibility of our approach and its applicability to di erent domains and to customised processes. We also demonstrate the capability of EXE-SPEM to model cloud-based executable processes. Furthermore, we demonstrate the added value of the process models and the process execution provenance data recorded by the SDaaS architecture. This data is used to automate the generation of safety cases argument fragments. Thus, reducing the development cost and time. Finally, the case study shows that we can integrate some existing tools and create new ones as activities used in process models. The proposed SDaaS reference architecture (combined with its modelling, scheduling and enactment capabilities) brings the benefits of the cloud to software development. It can potentially save software production cost and provide an accessible platform that supports collaborating teams (potentially across di erent locations). The executable process models support unified interpretation and execution of processes across team(s) members. In addition, the use of models provide managers with global awareness and can be utilised for quality assurance and process metrics analysis and improvement. We see the contributions provided in this thesis as a first step towards an alternative development method that uses the benefits of cloud and Model-Driven Engineering to overcome existing challenges and open new opportunities. However, there are several challenges that are outside the scope of this study which need to be addressed to allow full support of the SDaaS vision (e.g., supporting interactive workflows). The solutions provided in this thesis address only part of a bigger vision. There is also a need for empirical and usability studies to study the impact of the SDaaS architecture on both the produced products (in terms of quality, cost, time, etc.) and the participating stakeholders.