Next generation virtual datacenter management and monitoring framework

点击翻译按钮获取中文摘要

Cloud computing and virtualization technologies are dramatically changing how IT systems operate. What used to be a relatively static environment, with fixed physical nodes, has quickly transformed into a highlydynamic environment, where (clusters of) virtual machines (VMs) are programmatically provisioned, started, replicated, stopped and deprovisioned with cloud APIs. VMs have become the processes of the cloud OS, with short lifetimes and a rapid proliferation trend.****While the nature of data center operations has changed drastically, the management methodology of these (virtual) machines has not adapted appropriately. For example, common tasks, such as performance monitoring, compliance and security scans, and product discovery amongst others are carried out using repurposed versions of tools originally developed for managing physical systems. Unfortunately, this approach does not leverage new opportunities created by virtualization and as a result fails to deliver on the promise of high resource utilization, strong workload isolation, and predictable application performance.****We argue that realizing the potential of the modern data center requires redesigning the management framework from the ground up. This proposal is a joint effort between the University of Toronto and IBM to research a new management framework that leverages the virtualization layer to monitor, analyze, and actuate. A first thrust researches novel cluster monitoring infrastructure that leverages virtualization technology to extract detailed, comprehensive and consistent cluster and application information. A second thrust leverages our novel monitoring infrastructure to develop two applications that address the key challenges of VM placement and anomaly detection.**