Meeting latency targets for complex SOA-based transactions in data centers
Martin Johnsson, Brendan Jennings · 2016
Service-Oriented Architectures can experience a highly diversified and complex set of interactions between web services for them to compute a response to an incoming service request. Existing approaches based on modelling and control theory are good for strategic and long-term development of a data center resource dimensioning necessary to respond within the boundaries of Service Level Agreements. When operating at a scale of many thousands of incoming service requests per second, combined with a highly diversified set of a priori unknown service interactions, other more reactive approaches are required to cope with performance bottlenecks in a live production system. We have taken an experimental approach to look at how techniques inspired by SDN principles could potentially mitigate problems related to excessive response times while service requests are being processed. We conclude that significant improvements are possible and further observe, following the results from this study, the need to extend the scope of SDN to also incorporate programmability of application server nodes.