Dynamic partial reconfiguration (DPR) enabled FPGA-based Cloud architecture acts as a flexible and efficient shared environment to facilitates application support to users’ request at low cost. While on one hand we need to handle a variety of tasks, such as periodic or sporadic, deadline or non-deadline, high or low critical tasks from the point of producing correct results, on the other hand we are constrained to use untrusted FPGA-based application IP blocks procured from various third-party vendors, which may contain hardware Trojan horse (HTH) affecting throughput and reliability of the Cloud. We propose Trojan-aware processing of tasks by monitored execution of a task on different untrusted cores, and then one more execution is done upon detection of hardware Trojan effects. For this stringent scheduling environment, the proposed dynamic scheduling algorithm is also properly extended to guarantee successful recovery from Trojan effects for all accepted tasks. Experimental results show that our algorithm improves worst-case-response-time for all tasks including non-deadline tasks and achieves lower task rejection rate for the deadline tasks, through judicious non-uniform partitioning of FPGAs based on supported jobs and subsequent better resource utilization, compared to that for existing Trojan-aware scheduling techniques.

中文翻译：

---

在基于 FPGA 的云上使用硬件木马的恢复保证最小化 WCRT

支持动态部分重配置 (DPR) 的基于 FPGA 的云架构充当灵活高效的共享环境，以低成本促进对用户请求的应用支持。一方面，我们需要从产生正确结果的角度处理各种任务，例如周期性或零星、截止日期或非截止日期、高或低关键任务，另一方面，我们受限于使用不受信任的 FPGA-从各种第三方供应商采购的基于应用程序 IP 块，其中可能包含影响云吞吐量和可靠性的硬件特洛伊木马 (HTH)。我们建议通过在不同的不受信任的内核上监视任务的执行来对任务进行木马感知处理，然后在检测到硬件木马效应时再执行一次。对于这种严格的调度环境，所提出的动态调度算法也得到了适当的扩展，以保证所有接受的任务都能从木马效应中成功恢复。实验结果表明，我们的算法通过基于支持的作业和后续更好的资源利用率对 FPGA 进行明智的非均匀分区，提高了包括非截止日期任务在内的所有任务的最坏情况响应时间，并降低了截止日期任务的任务拒绝率。 ，与现有的木马感知调度技术相比。