Architectural-Space Exploration of Heterogeneous Reliability and Checkpointing Modes for Out-of-Order Superscalar Processors

Abstract: Reliability has emerged as a key topic of interest for researchers around the world to detect and/or mitigate the side effects of decreasing transistor sizes, such as soft errors. Traditional solutions, like DMR and TMR, incur significant area and power overheads, which might not always be applicable due to power restrictions. Therefore, we investigate alternative heterogeneous reliability modes that can be activated at run-time based on the system requirements, while reducing the power and area overheads of the processor. Our heterogeneous reliability modes are successful in reducing the processor vulnerability by 87% on average, with area and power overheads of 10% and 43%, respectively. To further enhance the design space of heterogeneous reliability, we investigate combinations of efficient compression techniques like Distributed Multi-threaded Checkpointing, Hash-based Incremental Checkpointing, and GNU zip, to reduce the storage requirements of data that are backed-up at an application checkpoint. We have successfully reduced checkpoint sizes by a factor ~6x by combining various state compression techniques. We use gem5 to implement and simulate the state compression techniques and the heterogeneous reliability modes discussed in this paper.