Towards a distributed computing framework for Fog

Fog computing paradigm has introduced the concept of processing data near the data source. Unlike the cloud, fog computing includes devices with highly varying resources such as heterogeneous computing power, battery, bandwidth, delay, and mobility. The existing distributed computing frameworks, however, have mainly focused on the cloud environment where resources are highly consolidated and stable. This paper presents Crystal, a distributed computing framework for fog. An application consisting of one or multiple Crystal instances offers distributed processing and computing while taking advantage of location transparency, self-healing, auto-scaling and mobility support. Our prototype implementation of MapReduce on Crystal shows benefits of fog computing — fault-tolerant distributed processing over heterogeneous, unreliable, fog nodes while reducing overall latency, thanks to the framework enabling processing close to the data source.