Improving the Quality Attributes of a Monolithic Messaging Gateway

All software communicates, either with the operating system, with other software running on the same machine, or over a network.Typically some middleware is used to facilitate this communication, providing protocol conversion between otherwise incompatible senders and recipients, as well as to handle buffering for optimized bandwidth usage. This sender > middleware > recipient model holds for all abstraction levels, from the physical link layer up to the application layer, with differences only in the details. In this thesis we focus on the application layer, in particular on the group of middleware called ``messaging gateways''. For the validation of our contributions, we use an existing, real-world messaging gateway deployed world-wide for forwarding mobile text messages. As these messages have a per message cost, this gateway also has a module for credit management, used when billing the senders for their traffic. Our overall goal is to find ways to improve the quality attributes of such a gateway, particularly concerning message delivery throughput and reliability.First, we wanted a better understanding of how the round-trip times for outgoing requests varied, in order to correctly detect abnormal delays. This resulted in a new variant of exponential smoothing which we used in a novel algorithm to detect anomalies. This algorithm was then validated in a case study, resulting in a log file analysis tool now used in production.Second, the requirements for high throughput and strong reliability in some ways contradict each other. Strong reliability requires messages to be replicated to one or more other nodes, resulting in extra processing and network traffic, which lowers the throughput. We addressed this conundrum by first writing a problem formulation on how the quality attributes of a messaging gateway would be affected by a multi-node configuration, resulting in a review of state of the art and state of practice for multi-node systems. Next, we developed a data replication algorithm, and validated it in a controlled experiment. Its proof-of-concept implementation showed that even in a geo-distributed configuration, replication throughput can scale with the number of nodes.Finally, in order to ensure we were solving the right problems going forward, we also performed an architecture analysis of the messaging gateway based on its quality requirements. From this exploratory case study, we deduced a somewhat unexpected plan for migrating the balance management module to a set of microservices, thereby providing higher throughput for most users of our messaging gateway.