异构多处理器系统中结合混合故障及其跨层传播的可靠性优化方法

Reliability optimization has always been a hot research field in real-time embedded systems. Most existing works of reliability optimization only focus on handling one type of two errors (i.e., soft and hard errors), and oftentimes ignore the cross-layer error propagation phenomenon and the effects of heterogeneous characteristics of tasks and processors on reliability. However, the fact is that soft and hard errors may both occur and propagate through different hardware layers in real-world computing systems, including heterogeneous multiprocessor real-time embedded systems. To address the above problems, this project aims to study the reliability-aware schemes for heterogeneous multiprocessor real-time embedded systems suffering both soft and hard errors. First, we investigate the cross-layer propagation mechanisms of soft and hard errors to build the accurate cross-layer models of soft and hard errors. Then, we propose a uniform metric to effectively evaluate system soft and hard-error reliabilities based on the cross-layer models. Using the uniform metric, we further estimate the availability of the multiprocessor system. Finally, by exploiting the heterogeneous characteristics of tasks and processors, we develop task allocation and scheduling algorithms to improve the system overall reliability and thus achieve a near-perpetual and reliable operation of the system. We will realize and validate the proposed algorithms on a hardware testbed. The results of this project are meant to be a valuable addition to the knowledge of design and verification of heterogeneous multiprocessor real-time embedded systems.

可靠性优化一直都是实时嵌入式系统领域的研究热点。现有的可靠性优化方法往往只考虑了单一的软故障或硬故障，同时大多忽视了故障的跨层传播现象以及任务和处理器的异构特性对可靠性的影响。但在真实系统中，软/硬故障均可能发生并会跨层传播，包括异构多处理器实时嵌入式系统。针对上述问题，本项目拟围绕异构多处理器实时嵌入式系统中结合软硬混合故障及其跨层传播现象的可靠性优化方法展开深入研究。首先，探索软/硬故障的跨层传播机制，以构建准确的软/硬故障跨层模型；然后，基于该模型提出软/硬故障可靠性的统一评价准则，以有效衡量两类可靠性，从而进一步评估多处理器系统的可用时间；最后，结合任务和处理器在可靠性方面的异构特性，设计任务分配和调度算法，以实现软/硬故障可靠性的综合优化，使系统长久可靠地运行。本项目将在硬件平台上实现并验证所提算法的有效性。本项目的研究成果将为异构多处理器实时嵌入式系统的设计和验证提供新思路。

可靠性优化一直都是实时嵌入式系统领域的研究热点。现有的可靠性优化方法往往只考虑了单一的软故障或硬故障，同时大多忽视了故障的跨层传播现象以及任务和处理器的异构特性对可靠性的影响。但在真实系统中，软/硬故障均可能发生并会跨层传播，包括异构多处理器实时嵌入式系统。针对上述问题，本项目围绕异构多处理器实时嵌入式系统中结合软硬混合故障及其跨层传播现象的可靠性优化方法展开了深入研究。..该项目取得的主要创新成果包括：（1）建立了一种新的可靠性模型来计算由软故障导致的MTFF和由硬故障导致的MTFF，并基于两类故障的MTFF，设计了多处理器系统可用时间的计算方法；（2）提出了面向CPU多核处理器的任务调度算法，通过反馈控制来决定系统的负载量和任务的备份以最大化系统的可靠性；（3）提出了面向CPU-GPU异构多核处理器的任务调度算法，通过任务映射、任务动态迁移和动态电压调整技术，来平衡各个内核之间的磨损状态，以在峰值温度和实时约束下最大化系统可靠性；（4）设计了离线-在线混合式任务调度框架，根据处理器的负载、温度、能耗和磨损状态交替采用离线和在线算法以提高系统可用性。..本项目的研究成果将来可应用于异构多处理器实时嵌入式系统，能够有效降低异构多处理器错误率、减缓异构多处理器老化效应、提高异构多处理器任务调度算法的效率、增强异构多处理器系统可靠性，在嵌入式系统研究领域具有十分重要的理论意义，在物联网行业也具有较为广泛的应用前景。