With the rapid development of virtualization techniques, modern data centers move into a new era of cloud in recent years. Despite numerous advantages such as high resource utilization and rapid service scalability, current virtualization techniques don't guarantee perfect performance isolation among virtual machines sharing the physical machine, which may lead to unstable and unpredictable user-perceived application performance in clouds. Therefore, understanding and modeling performance interference among collocated applications is of utmost importance. However, the hypervisor and guest OSes usually run independent resource schedulers and are invisible into each other, thereby making accurately characterizing performance interference a non-trivial work.In this paper, we first present a comprehensive experimental study on performance interference of different combinations of benchmarks, observing that virtual CPU floating overhead between multiple physical CPUs, and VMEXITs, i.e., the control transitions between the hypervisor and VMs, constitute the key source of performance interference. In order to characterize the performance interference effects, we measure both the application-level and VM-level characteristics from the collocated applications and then build a novel interference prediction framework based on kernel canonical correlation analysis. Our evaluations first show the practicability of KCCA in finding reliable correlation, and further confirm the high accuracy and great applicability of our interference model with a low prediction error of no more than 7.9%.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.