By Hisham Muhammad, Kong Inc.
Typical userspace process monitoring tools usually show traditional metrics such as CPU and memory percentages, but modern hardware provides much more data.
Performance monitoring nowadays needs to consider these metrics, such as instructions-per-cycle and cache usage, as they paint a better of hardware utilization, taking into account superscalar processing and memory hierarchy.
The kernel already makes this data available to userspace, but it’s currenly used only in advanced tools such as systemtap and perf-events.
This talk will present the new hardware performance monitoring metrics introduced in htop 3.0, hoping to bring these measurements to a wider audience.