Cloud Platform Performance Evaluation Using Multi-level Execution Tracing

Performance evaluation is always an important issue in cloud environment. Problems can arise at different layers of the system with several different reasons and that is why detecting and then identifying the origin of a problem can be a difficult task. This research proposes a multi-level execution trace-based approach to detect performance issues in cloud services. The method works by collecting run-time data from several layers including user-space, kernel space and virtualization layer, correlating, modeling and analyzing in order to detect abnormal performance behavior. We have tested our method in virtual machine migration scenarios where it is difficult to collect trace data. Our results demonstrate that the proposed method can detect service inefficiency issues caused from various service and system level reasons.