With the rise in the adaptation of edge computing frameworks for application deployment, one should decide where to place the enormous amounts of data generated at the edge to provide satisfactory services. Some edge applications like augmented reality games have co-located end-users which provide an opportunity to use location and network proximity as measures to identify the best storage nodes. However, given the resource constraints of heterogeneous edge server nodes, data placement algorithms should consider the storage capacity, fan-in and fan-out limits to ensure low-latency services. In this paper, we discuss three data placement strategies (distance, latency, and spatial) that consider different factors like location, network latency, storage capacity, and fan in/out distributions with dynamic replication of read-only data. Based on our simulation and emulation experiments, distance and latency-based strategies are best suited for sparse edge environments and the spatial for dense edge environments.
View on arXiv