Programming Methodology for Multi-Core Concurrency and Adaptation

今天的软件密集型系统范围从移动设备到嵌入软件的产品，再到全球联网系统。随着我们对软件的依赖程度越来越高，软件的可靠性越来越受到社会的关注。我们的长期目标是“正确的软件”的理想。该项目的贡献是通过解决两个问题：并发性和适应性，使软件的设计和发展更加可靠。 由于多核处理器的出现，越来越多的软件被编写用于并发执行。通过支持并发执行，多核处理器以更低的功耗提供更高的性能，这具有扩展移动设备的实用性、减少冷却需求和降低运营成本等好处。并发性的正确性是出了名的难以实现。“原子操作”理论通过允许逐步引入并发来解决这一问题。我们计划为具有原子操作的面向对象编程语言开发高效的实现技术。 软件系统也是“多面性的”，并在不断发展。身份验证、容错、隐私、功耗、分发和并发等问题往往会影响程序的许多部分，并且很难分离出来。静态“方面”允许在开发过程中分阶段实现这些关注点。需要实现额外的功能，并且需要在发展过程中改进关注点。已部署并持续运行的操作系统必须为此进行扩展和修改。动态的“混合”允许这样的适应。方面和混合的正确性是一个悬而未决的问题。我们计划定义一个“行为保持”方面和混合的理论，并让一个“智能”的编译器来分析这一点。例如，这样的分析允许检测软件更新是否会破坏现有功能。

Today's software-intensive systems range from mobile devices to products with embedded software to worldwide-networked systems. As we become increasingly dependent on software, the reliability of software is of increasing concern to society. Our long-term goal is the ideal of "correct software". The contributions of this project are making the design and evolution of software more reliable by addressing two issues: concurrency and adaptation. Software is more and more written for concurrent execution due to the emergence of multi-core processors. By supporting concurrent execution, multi-core processors offer higher performance at lower power consumption, which has benefits like extended utility of mobile devices, reduced need for cooling, and lower operating cost. Correctness of concurrency is notoriously difficult to achieve. The theory of "atomic actions" addresses this by allowing a stepwise introduction of concurrency. We plan to develop efficient implementation techniques for an object-oriented programming language with atomic actions. Software systems are also "multi-faceted" and constantly evolving. Concerns, like authentication, fault- tolerance, privacy, power consumption, distribution, and concurrency tend to affect many parts of a program and are difficult to isolate. Static "aspects" allow such concerns to be implemented during the development in stages. Additional functionality needs to be implemented and concerns need to be improved during evolution. Deployed and continuously operating systems have to be extended and modified for this. Dynamic "mix-ins" allow such adaptation. The correctness of aspects and mix-ins is an open issue. We plan to define a theory of "behaviour-preserving" aspects and mix-ins and have an "intelligent" compiler analyze this. Such an analysis, for example, allows to detect if a software update would break existing functionality.