DRMaestro: orchestrating disaggregated resources on virtualized data-centers

Abstract Modern applications demand resources at an unprecedented level. In this sense, data-centers are required to scale efficiently cope with such demand. Resource disaggregation has the potential improve resource-efficiency by allowing deployment of workloads in more flexible ways. Therefore, industry is shifting towards disaggregated architectures, which enables new ways structure hardware data centers. However, determining best performing resource provisioning a complicated task. The optimality allocation center depends on its topology and workload collocation. This paper presents DRMaestro , framework orchestrate transparently from applications. uses novel flow-network model determine optimal placement multiple phases while employing best-efforts preventing performance interference. We first evaluate impact regarding additional network requirements under higher load. results show that for some minimal, but other ones can suffer up 80% slowdown transfer part. After that, we via real prototype Kubernetes trace-driven simulation. reduce total job makespan speedup ?1.20x decrease QoS violation ?2.64x comparing another orchestrator does not support disaggregation.