perf-rt: Performance Measurement Through Event-Based Data Sampling
What is it?
The more precisely performance is determined, the less precisely the performance is known (Heisenberg's Uncertainty Principle on measurement). The relationship between system performance and the act of measuring system performance is dependent. What this means is that, in reality, the act of measuring performance changes performance.
Perf-rt is an application utility that provides real-time performance data while reducing the uncertainty inherent in the act of measurement. Perf-rt measures performance by event-based data sampling, which is less intrusive than the incumbent periodic (time-domain sampling).
What are its benefits?
Perf-rt will improve rt-kernel performance measurability in production environments, enabling crisp performance characterization under genuine real-time workloads. Perf-rt will enable superior application tuning by providing more accurate, targeted, and relevant data collection. It will also reduce the need to create artificial and synthetic workloads for tuning system performance and identifying bottlenecks through a minimally invasive event-driven performance sampling algorithm.
When will it be used?
Perf-rt is expected to be used by customers who seek to improve system performance through tuning RHEL and/or native applications running on their systems.
These customers might be from the financial sector, seeking to improve application performance through increased system utilization and bottleneck analysis. Some examples are trading desks, stock exchanges, banking operations and transactions, and federal customers who have mission-specific exotic applications that would benefit from the superior system-tuning capabilities of on-the-fly performance measurements.