Efficient cloud tracing: From very high level to very low level

With the increase of cloud infrastructure complexity, the origin of service deterioration is difficult to detect because issues may occur at the different layer of the system. We propose a multi-layer tracing approach to gather all the relevant information needed for a full workflow analysis. The idea is to collect trace events from all the cloud nodes to follow users' requests from the cloud interface to their execution on the hardware. Our approach involves tracing OpenStack's interfaces, the virtualization layer, and the host kernel space to perform analysis and show abnormal tasks and the main causes of latency or failures in the system. Experimental results about virtual machines live migration confirm that we are able to analyse services efficiency by locating platforms' weakest links.