Analytics and Visualization in Openstack (AVOS)

Abstract

Openstack and other virtualized cloud computing systems are vulnerable to performance faults and anomalies [1] due to various reasons, such as resource exhaustion, software flaws, and hardware crashes. Currently, Openstack Ceilometer, the telemetry module of Openstack, is only used by system administrators and developers for custom solutions. There exist no means of monitoring individual performances of virtual machines (VMs) within a tenant as well as no system employs predictive analytics on the data stream provided by Ceilometer. In this project, we present a novel Analytics and Visualizations in Openstack (AVOS) system that provides black-box anomaly detections for VMs, visualization for cloud components and predictive analytics to monitor VM’s performance and state over time. AVOS provides a web based interface for users and system administrators to monitor their resources running within Openstack Cloud and get notified for any anomaly detected through our anomaly detection model. We have implemented AVOS on top of Openstack Cloud, exploiting Openstack Ceilometer APIs to acquire real-time data and tested it with common workloads (such as video streaming server and distributed Hadoop job). Experimental Results showed that AVOS can learn the system’s behavior over time and can efficiently report performance anomalies with only system level metrics, having low overhead on overall cloud infrastructure.