Cooperative cross-layer protection for resource constrained mobile multimedia systems

With rapid advances of technology and wide deployment of wireless communication, mobile embedded systems are becoming popular. However, system failures increase significantly due to increasing complexity of integration and increasing error rates as technology scales, and thus reliability is becoming a critical concern. Incorporating reliability in resource-limited mobile embedded systems poses significant challenges due to high overheads of conventional redundancy techniques in terms of performance, power, cost, etc. For example, error correction codes for cache protection incur more than 22% power overhead, and triple modular redundancy techniques for logic component protection incur 200% cost of area and power without optimization. Using the observation that error-awareness such as error-tolerance, error-resilience, and error-concealment can be exploited to enhance system properties, this thesis proposes a cross-layer methodology for mobile embedded systems with minimal overheads by exploiting error-awareness across system abstraction layers in a cooperative manner. Previously proposed cross-layer methods have focused on power, performance, QoS, and timing issues rather than reliability issues. This thesis investigates errors and error control schemes across system abstraction layers, and presents error-aware, cross-layer approaches that exploit existing techniques to mitigate the impact of the different classes of errors, and further exploit errors actively for maximal resource savings. This thesis demonstrates the effectiveness of our cooperative, cross-layer methodology in several ways for mobile embedded systems. We have investigated PPC (Partially Protected Caches) architectures that exploit error-tolerance and vulnerability of applications at the application layer to combat soft errors at the hardware layer. We developed EAVE (Error-Aware Video Encoding) that proposes active error exploitation for maximal energy reduction by intentionally dropping video frames at the middleware layer and managing the quality with error-resilience against network errors. Finally, we developed CC-PROTECT (Cooperative, Cross-layer Protection) that jointly orchestrates error detection schemes at the hardware layer, frame dropping and forward error correction at the middleware layer, and error-aware video encoding at the application layer. Our cross-layer approaches significantly reduce resources such as power and area cost for resource-constrained embedded systems, open up an expanded tradeoff space for multi-dimensional constraints, and eventually enable system designers to explore feasible solutions satisfying maximal reliability with minimal overheads of power and performance.