forked from sfu-dis/pibench
-
Notifications
You must be signed in to change notification settings - Fork 0
Tail Latency
Lucas Lersch edited this page Mar 19, 2020
·
2 revisions
PiBench can collect the latency of percentage of the total amount of request with the option --latency_sampling=[0.0, 1.0].
This is the probability of the time of individual requests being measured.
A higher probability will result in more precise latency measurements, but also higher overhead.
The user is encouraged to try different percentages and compare latency and throughput numbers.
At the end of the execution the percentiles of the collected measurements is printed in nanoseconds (as seen above).
Contents
- Home
- Building
- Running
- Intel PCM
- OpenMP
- Skipping Load Phase
- Tail Latency
- Internals
- Key Generator
- Value Generator
- Op Generator