A distributed Cloud Continuum observability architecture to support Next Generation Networks

Cloud Computing is a widespread deployment paradigm, and to guarantee high performance and high availability, it needs to be constantly monitored. As most of the deployed services cannot be modified at a source code level to instrument them for exposing monitoring data, it is used their output instead. This technique is called Observability. The rise in popularity of new technologies and standards, such as IoT and 5G networks, increased the need to have computational power near the edge of the network, creating an evolution of Cloud Computing, which is Cloud Continuum. Cloud Continuum use cases have different, and tighter, requirements with respect to Cloud Computing, thus centralized observability solutions, popular in the latter deployments, cannot fulfill the needs of the former, due to their centralized nature. This work presents a distributed observability agent which reduces the response times and the network traffic. It is tested against two major centralized observability stacks, that is, the Prometheus and Grafana pair, and the Elastic suite, for resource usage (CPU, RAM, and bandwidth), and time to alert. As the agent leverages WebAssembly modules to customize functions available in its workflows, a set of tests compares this language with other popular languages, which are Kotlin, JavaScript, and Rust, for resource usage (CPU, RAM, container image or module size), performance, and startup time. Both suites of tests confirm the validity of the proposed solution: the agent, on average, requires the same resources as the other observability stacks, but it uses less bandwidth and it has a faster reaction time; WASM performance is equivalent to a compiled language, such as Rust, while its module size and startup time are considerably lower, allowing to respond quicker to request peaks.