Load Balancing for Minimizing Deadline Misses and Total Runtime for Connected Car Systems in Fog Computing

Cloud computing provides a pool of highly available resources for applications to use to offload their tasks, but new applications such as coordinated lane-change assistance used in connected car systems have strict timing requirements that cannot be met by offloading tasks only to the cloud. Fog computing reduces latency by bringing the computation from remote datacenters to local fog servers, which are connected in close proximity to clients. Although fog computing lowers the latency for transferring data, load balancing among fog servers still needs to be addressed for better timing performance. The challenges include a large number of tasks, mobility of the clients, and heterogeneity of the fog servers. In this paper we use connected car systems as a motivating application, and first show that we can utilize mobility patterns of vehicles to perform periodic load balancing in fog servers.We then present a task model that solves the scheduling problem at the server level instead of device level. And finally we formulate a load balancing optimization problem for minimizing deadline misses and total runtime for connected car systems in fog computing. We show that it outperforms some common heuristics such as weighted round-robin, active monitoring, and throttled load balancer.