University of Twente Student Theses

Login

A lightweight hardware architecture for intermittent computing

Rayo Torres Rodriguez, H. (2019) A lightweight hardware architecture for intermittent computing.

This is the latest version of this item.

[img] PDF
1MB
Abstract:The pervasiveness nature of IoT devices and recent developments on energy-harvesting technologies have enabled the development of a novel brand of battery-less devices whose operation relies entirely on energy harvested from the environment. Unfortunately, ambient energy leads these devices to experience frequent power failures, making data consistency, and forward progress, one of the biggest challenges to overcome. Most of the research on this domain has focused on software and compiler-based approaches to allow existing off-the-shelf processors to operate intermittently. Nevertheless, most of them present abstractions difficult to use for non-expert programmers, incur in high run-time overhead, or limit programmability features common to low-level programming languages. To address these shortcomings, we propose Qwark, a hardware/ software solution which utilizes simple yet efficient memory management techniques to transparently solve the problems introduced by intermittency with little extra overhead and no programmer intervention. Qwark continuously tracks memory accesses and uses address translation to isolate harmful accesses, creating a log in an isolated memory segment, effectively preventing the system from reaching inconsistent states due to uncommitted non-volatile memory writes. Furthermore, during run-time, Qwark promotes variable-sized segments of the stack to non-volatile memory to enforce a worst-case checkpoint time, using address translation to link volatile and non-volatile segments, reducing run-time and power consumption. This work also presents the first hardware solution targetting the MSP430 architecture, to provide a fair performance assessment against current software-based solutions. Qwark is evaluated running a set of benchmarks, showing significant improvements over existing state of the art software-based implementations.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Embedded Systems MSc (60331)
Link to this item:http://purl.utwente.nl/essays/79568
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page