You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi @JaykumarPatel4802, thanks for asking! The RAPL MSRs update approximately every millisecond, but the actual measurement period is not precisely 1 ms, leading to variations in sample periods.
In fact, RAPL provides energy consumption data rather than power consumption, and lacks timestamps for each counter update [1]. This limitation makes it impractical to achieve rates exceeding 20 samples/s if the goal is to keep systematic error below 5%. Also, continuously polling the RAPL registers not only consumes processor core resources but also distorts the accuracy of the measurements [2], which might increase the cost of energy tracing.
Hope this explanation helps a bit. Feel free to reach out if you have any other questions!
[1] Khan, Kashif Nizam, et al. "RAPL in Action: Experiences in Using RAPL for Power measurements." ACM Transactions on Modeling and Performance Evaluation of Computing Systems (TOMPECS) 3.2 (2018): 1-26.
[2] Hackenberg, Daniel, et al. "Power measurement techniques on standard compute nodes: A quantitative comparison." 2013 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS). IEEE, 2013.
tracer.py logs a critical error if rapl_interval_sec < 0.05.
The text was updated successfully, but these errors were encountered: