In a virtualization system, problems like I/O and scheduling delay sometimes occur on guests because those operations of guests and the host will compete by sharing I/O devices or CPU cores. However, if you just look into only guest's trace data, it will be difficult to analyze the problems. So, we are developing "Integrated trace" system which allows us to analyze trace data of all guests and a host by merging data in chronological order. Our proposal is to use TSC for merging and the concept was reviewed by the community, and we found that there are two problems: TSC offset changing and difference of TSC between multiple CPUs.
In this presentation, we report current status of Integrated trace, share the problems using TSC in detail, and explain how to approach for that. This presentation will be a help for troubleshooting on virtualized mission-critical systems or cloud systems.