The rapid growth in numbers and computational capacities of IoT/mobile devices has led to the emergence of FemtoClouds: systems that extend Edge resource to pools of IoT/mobile devices. By virtue of relying on clusters of smart IoT/mobile devices for offloading, FemtoCloud resources are highly heterogenous, dynamic, and footprint-sensitive: highly sensitive to workloads whose footprints hinder the normal functioning of the device. This imposes several requirements that existing offloading mechanisms do not fulfill – such as architecture-awareness, maximizing utilization, and light-weightness. This project introduces FemtoCOF, an offloading framework designed from the ground up with FemtoCloud requirements in mind. Based on LLVM as a cross-compiler, FemtoCOF precompiles code for each architecture in the FemtoCloud system and adds a machine-dependent-optimization layer via extra LLVM passes. Finally, we develop a proof-of-concept prototype for our offloading framework, and provide preliminary evaluation of its faithfulness to FemtoCloud requirements in comparison to existing solutions.
