Edge-Facilitated Wireless Distributed Computing

We consider a mobile computing application where a group of mobile users, which all connect to a server at the edge of a mobile network, each has a request to process using a dataset. To run this application distributedly, we propose a wireless distributed computing framework, motivated by commonly used structures like MapReduce, where the overall computation is decomposed into computing a set of "Map" and "Reduce functions distributedly across all the participating users who has locally stored parts of the dataset prior to the computation. For such framework, we propose a coded scheme, named "Coded Distributed Computing" (CDC), to demonstrate that for a given computation load of the Map phase r (i.e., in average each Map function is evaluated by r users), coding can significantly reduce the communication load on the uplink towards the edge server and the communication load on the downlink towards the users, by a factor of r and r + 1 respectively.