Measurements in the edge-cloud continuum: network metrics and energy consumption

Edge computing is an emerging network paradigm based on the idea of moving computational and storage resources closer to the end users. This type of architecture can bring a variety of benefits compared to the traditional cloud-based one, such as lower latency, higher throughput, increased privacy, and reduced congestion of the Internet core. However, making an effective use of edge computing requires to monitor the performance of the network, e.g. to take appropriate decisions about if and where computation should be offloaded, or which server in the edge-cloud continuum is more suitable for a given operation. Most of the existing network measurement methods and tools are not specifically designed to operate in an edge computing scenario. For this reason, mechanisms aimed at collecting network metrics in an edge environment have been first designed and then used to collect experimental data in a realistic testbed. Network measurements can also be useful at design time, e.g. to evaluate different edge/cloud solutions by means of trace-driven simulations, as discussed in this thesis. The energy needed by client devices to communicate with edge or cloud resources is another important aspect, since such devices, which include smartphones and IoT nodes, are generally battery-operated. To better understand how the edge computing paradigm impacts the energy needed to communicate, an analytical model of a request-response communication scheme has been defined. The model highlights that the improved latency of an edge server, compared to a cloud one, can reduce the energy needed by clients. Energy savings are particularly significant when communication takes place according to a connection-oriented protocol. This thesis also looks at the path between client nodes and cloud resources from a purely topological perspective. Traceroute is the most commonly used tool, not only for network diagnostics, but also for discovering all the nodes towards a server. We evaluated the discovery capability of three variations of TCP-based traceroute. The first version is the classical one and uses SYN segments as probes. The other two versions operate on a connection already established with a server and use DATA and ACK segments as probes. This is done to possibly bypass traceroute suppression mechanisms or firewalls. Experimental results show that using different types of probes is useful to obtain a richer view of the path towards network resources.