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
Firstly, thank you and all the contributors for this awesome tool.
Goal: to simulate SQL queries from an application, where users may not wait for a query "A" to finish, and immediately trigger one more query "B", producing even more (IO-bound) load on the database (DB).
Questions:
Would running with --threads=1 and executing a slow SQL query mean that the next query following would be waiting on it? That is, would DB effectively push back and back-pressure against any further load? The queue_time is included in the elapsed time measurement, which is good. But even more realistic would be if a user/thread would be able to pile on additional load to the DB at will.
I think the answer to the above lies in the code on these 2 lines (I'm bit rusty on pthreads):
Would you use --threads=20 --rate=20 to simulate 20 concurrent users, each producing 1 QPS? Does one have to also consider number of CPU cores (and/or hyperthreads), or anything else, when choosing these flags?
Using sysbench 1.1.0-2ca9e3f (using bundled LuaJIT 2.1.0-beta3).
The text was updated successfully, but these errors were encountered:
oatmealb
changed the title
[Question] Why does benchmarking a single SELECT 1 returns twice as many queries as the transactions?
[Question] Could running with --threads=1 and --rate=N cause backpressure, if the database can't keep up with the rate?
Sep 15, 2023
oatmealb
changed the title
[Question] Could running with --threads=1 and --rate=N cause backpressure, if the database can't keep up with the rate?
[Question] Would running with --threads=1 and --rate=N cause backpressure, if the database can't keep up with the rate?
Sep 15, 2023
Firstly, thank you and all the contributors for this awesome tool.
Goal: to simulate SQL queries from an application, where users may not wait for a query "A" to finish, and immediately trigger one more query "B", producing even more (IO-bound) load on the database (DB).
Questions:
Would running with
--threads=1
and executing a slow SQL query mean that the next query following would be waiting on it? That is, would DB effectively push back and back-pressure against any further load? Thequeue_time
is included in theelapsed
time measurement, which is good. But even more realistic would be if a user/thread would be able to pile on additional load to the DB at will.I think the answer to the above lies in the code on these 2 lines (I'm bit rusty on pthreads):
Would you use
--threads=20 --rate=20
to simulate 20 concurrent users, each producing 1 QPS? Does one have to also consider number of CPU cores (and/or hyperthreads), or anything else, when choosing these flags?Using sysbench 1.1.0-2ca9e3f (using bundled LuaJIT 2.1.0-beta3).
The text was updated successfully, but these errors were encountered: