Adaptive Replica Selection in Mobile Edge Environments

Mobile Edge Computing (MEC) is a paradigm that aims to bring cloud services closer to mobile clients, effectively reducing latency and saving backbone bandwidth. As in cloud environments, many applications make use of replication to enhance their quality of service. However, here, data generated by the mobile devices is usually kept near its source, and can have multiple replicas scattered through the network (e.g., on the mobile devices or on edge servers). When requesting data, replica selection can have a significant impact in multiple aspects of a system, e.g., load balancing, throughput, or energy efficiency. Thus, the possible herd behavior combined with the unreliable wireless communication channels can cause systems to under-perform. In this paper, we propose Mecerra, a replica ranking algorithm tailored for the characteristics of MEC environments. Additionally, we detail Wasabi, a flexible replica ranking framework that also handles the management of system metrics. We implement Mecerra in Wasabi, and integrate it into a data storage system for edge networks, building an adaptive replica selection scheme. We use the resulting system to evaluate our proposal and compare it against related work. Results show that Mecerra is able to greatly increase the probability of finding the best replica, and Wasabi provides low overhead.