Key-Value Stores in the Edge: An Implementation Using Replicated State Machines

Edge computing not only brings computation closer to data sources, but also moves data closer to clients, making it a more suitable approach than cloud computing for applications where low latency is crucial. However, designing a performant storage system to run in an edge network is a challenging task due to several characteristics of edge networks, such as partitions in the network and devices with limited resources. In this paper, we introduce the pire-store, a distributed key value store based on replicated state machines that will be used in edge networks. pire-store offers low latency, runs distributed algorithms for storage operations, can be enabled in partial mesh networks (PMNs), and utilizes eventual consistency. By running multiple handlers, the pire-store is designed to be robust against the increasing number of simultaneous users. We implement the proposed design using Python and then compare it with etcd, Apache ZooKeeper on a Raspberry Pi cluster. We compare the products for each storage operation and for realistic cases. The results show that the proposed design performs approximately 30% faster than etcd and Apache Zookeeper as the number of simultaneous users increases.