Enabling Job-aware scheduling on Kubernetes clusters

Enabling Job-aware scheduling on Kubernetes clusters. Over the last decade we have noticed a deep change in the development of web based application; developers now can benefits of the massive computational power of data-centers and cloud environment by leveraging on microservices and containerization. At the mean time, while reducing the effort for developers, it increased the complexity for Cloud providers and in particular for Cloud orchestration platforms. This is why over the last decade a big effort has been put in research for solutions aiming to reduce the complexity of this development framework (profiling is indeed one of these previously mentioned research topic). The outline of the thesis will be structured as follows: [CH1-Introduction] This chapter provide a brief introduction on both the current scenario for microservice orchestration and the reasons behind the decision to develop this thesis [CH2-Kubernetes] This chapter is an introduction to Kubernetes as a container orchestrator, focusing mainly on the features that will be exploited in the work of this thesis, providing a general background for the concepts that will be presented afterwards [CH3-Profiling in cloud environment] This chapter describes what's the state of the art for what concerns the profiling of microservice. It briefly introduce the most relevant papers which already faced this research topic. This thesis is meant to be a part of the Liqo project, so the final section of this chapter will introduce it [CH4-Job profiling design] This chapter describes the design behind the profiler developed in this thesis, analyzing first microservices common behaviour in production datacenter, moving then to the actual logical design of the system [CH5-Job profiling implementation] This chapter starts from the logical implementation described in the previous chapter and provides a more practical description of the actual implementation of the algorithm [CH6-Experimental evaluation] This chapter evaluates the performance of the profiling algorithm described previously in a real scenario [CH7-Conclusion and future work] This final chapter starts from the results obtained in the previous chapter and evaluates them, providing a critical analysis of the algorithm and some possible improvements