Nonintrusive Measurement on Temporal and Spatial Features of Microservice Inferences

Abstract The high flexibility of microservice architecture provides notable divergence among the internal software stack within the same application. Microservice-based application are commonly deployed in data center from users yet they have no cue where or what existly are provided from the service provider. In this case, there could exist those providers could be able to replace the internal software without noticing, making a contract level fault and offloading risks to users. To better profile the microservice, we propose a framework that provides non-intrusive measurements on microservice inference, or MEME, to detect fraudulent behaviors of cloud service providers on microservice-based applications. We design MEME using performance portrait and fast Fourier transform (FFT). We model the microservice-based application with a set of cohorts and use FFT to obtain the signal formed by the main frequency components of average response time. Our model represents the performance portrait of the microservice-based application. The performance portrait is similar to a fingerprint that Nonintrusive Measurement of Microservice Inferences carries internal software information. In our experiments, we take a two-tier microservice-based application containing databases as an example. Empirical results show that MEME can provide a distinguishable profile of the performance portrait of data services in a temporal and spatial manner, which allows us to identify the software type.