HerdMonitor: Monitoring Live Migrating Containers in Cloud Environments

Cloud computing uses pools of virtual machines to provide shared computing resources. Provisioning and management of these resources are generally done using statistical algorithms to help decide how to better utilize available compute power; recently, this has been performed mostly by using live migration of virtual machines. Nowadays containers deliver the flexibility to handle many software environments and tasks in a lighter-weight virtualization scheme, providing a more agile alternative to virtual machines for certain applications. Application checkpointing coupled with a container manager allows the live migration of a container. In this work, we present Herd-Monitor, a resource monitoring system to gather performance metrics of containers and their respective compute hosts. This data can be used to analyze the utilization of available resources, characterize workloads, and develop forecast models. With the analysis of the data provided by the monitoring system, we can fine-tune resource provisioning policies, with the end goal of constructing a cloud resource provisioning platform that enables an improvement in usage and execution of container workloads through the use of live migration. The design, development, performance evaluation and characterization of our lightweight resource monitoring tool that enables offline and real-time resource utilization analysis of live migration workloads as well as the performance impact on their hosts are presented.