Building Error Resilient Software on Next Generation Computing Platforms

随着计算机系统变得更小、更快，其可靠性成为一个重要的问题。实现高可靠性的一种方法是设计具有许多冗余组件的系统，以便即使其中一些组件发生故障，它也可以继续运行。然而，这种方法会带来巨大的电力和能源消耗成本，这在智能手机和数据中心等下一代计算机系统中非常重要。因此，我们需要新的方法来确保不依赖于冗余硬件的未来计算机系统的可靠性。*我们建议通过构建对硬件故障具有弹性的软件，在应用软件级别容忍硬件错误。这样做的主要理由是，很少有错误会在系统堆栈中向上影响应用程序，即使是那些影响应用程序的错误也不一定对应用程序重要。因此，在不增加大量冗余的情况下，保护软件应用程序免受(大多数)硬件故障的影响要高效得多。然而，主要的挑战是，程序员很难在软件层面上推理硬件故障，并在不花费大量努力的情况下保护他们的应用程序。我们的目标是完全自动化使应用程序具有容错能力并对其进行推理的过程，而不会给程序员带来负担。这需要用于程序(静态)分析、正式方法和经验评估的新技术，这将是本提案的重点。*通过利用本提案中的研究，下一代计算机系统将能够提供更高的性能和能效，即使在可靠运行的情况下也是如此。这反过来可以导致更多人采用移动计算和智能手机，以及更节能的互联网数据中心。这些都是经济增长的驱动力，因此，本提案中的研究对这些领域具有重大的潜在影响。此外，在该项目中培训的HQP将获得设计和验证可靠计算机系统的重要技能，随着计算机系统变得越来越复杂，这一技能变得更加重要。最后，拟议的研究可以为下一代计算机系统带来显著的能源节约，预计到2020年，下一代计算机系统将占据总能源消耗成本的很大一部分，从而使整个社会受益。

As computer systems become smaller and faster, their reliability becomes an important concern. One way of achieving high reliability is to design the system with many redundant components, so that it can continue operating even if some of them fail. However, this approach entails significant costs in power and energy consumption, which are very important in next generation computer systems such as smartphones and data centers. Therefore, we need new methods to ensure the reliability of future computer systems that does not rely upon redundant hardware.***We propose to tolerate hardware errors at the application software level, by building software that is resilient to hardware faults.  The main rationale for doing so is that few errors make their way up the system stack to affect the application, and even the ones that do are not necessarily important to the application. Therefore, it is much more efficient to protect software applications from (most) hardware faults, without adding significant amounts of redundancy. The main challenge however is that it is difficult for programmers to reason about hardware faults at the software level, and to protect their applications without expending a huge amount of effort. Our goal is to completely automate the process of making applications error resilient and reasoning about them, without burdening the programmer. This requires novel techniques for program (static) analysis, formal methods, and empirical evaluations, which this proposal will focus on. ***By leveraging the research in this proposal, next generation computer systems will be able to deliver higher performance and energy efficiency, even while operating reliably. This in turn can lead to higher adoption of mobile computing and smart phones, as well as more energy-efficient internet data centers. These are the drivers of economic growth, and as such, the research in this proposal has significant potential impact on these domains. Further, the HQP trained in this project will acquire important skills to design and validate reliable computer systems, which is becoming more important as computer systems are becoming more complex. Finally, the research proposed can lead to significant energy savings for next generation computer systems, which are predicted to occupy a substantial fraction of the total energy consumption costs by 2020, thus benefiting society as a whole.**