Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Reported rate can be higher than actual when latency is high #80

Open
tgrabiec opened this issue Jan 6, 2022 · 0 comments · May be fixed by #93
Open

Reported rate can be higher than actual when latency is high #80

tgrabiec opened this issue Jan 6, 2022 · 0 comments · May be fixed by #93
Labels

Comments

@tgrabiec
Copy link

tgrabiec commented Jan 6, 2022

In the following output, reported rate is close to 4k rows/s but the actual rate is half that.

scylla-bench executes requests concurrently according to the configured concurrency and emits a partial result from each concurrent fiber if more than a second has elapsed. Those partial results are then summed up together and assumed to have been produced over a period of a second. When latency is high, the period is actually larger than a second, and actual rate lower.

time                operations/s   rows/s   errors  max              99.9th           99th             95th             90th             median           mean             
2.416392808s                4027     4027        0  1.023410175s     1.013972991s     1.007681535s     999.817215ms     996.147199ms     966.262783ms     747.089756ms     
4.448323544s                3485     3485        0  1.035468799s     1.031274495s     1.020788735s     1.014497279s     1.010827263s     993.525759ms     993.77632ms      
6.339966315s                3630     3630        0  1.040711679s     1.029177343s     1.019740159s     1.010302975s     1.002438655s     974.651391ms     964.700896ms     
8.189235084s                3734     3734        0  1.029177343s     1.024458751s     1.013972991s     1.004535807s     998.768639ms     923.271167ms     929.854113ms     

@tgrabiec tgrabiec added the bug label Jan 6, 2022
@dkropachev dkropachev linked a pull request Feb 28, 2022 that will close this issue
dkropachev added a commit to dkropachev/scylla-bench that referenced this issue Feb 28, 2022
Fix following things:
1. Get rid of 2 histograms per thread to reduce memory usage
2. Make latency printing non-blocking
3. Reorganize code
4. Get rid of histogram merging

fixes scylladb#80, scylladb#49, scylladb#73, scylladb#69
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Projects
None yet
Development

Successfully merging a pull request may close this issue.

1 participant