Good afternoon, i'm a new joiner, all apologies if i break any rules by posting my first post here.
My mean concern is looking for study, about turboboost limitation.
Let's say i have turboboost activated on my server and that usually on a 16cores cpu (model name : Intel(R) Xeon(R) CPU E5-2690 0 @ 2.90GHz
),
- only 5 or 6 are used 100% with polling process which are binded on distinct cores (one process per core, stuck at 100%cpu (while true do.... style)
- cstate enforced to 0 in bios and linux boot cmdline (intel_idle.max_cstate=0 processor.max_cstate=0 idle=poll)
If i add more processes, in this context cpupower monitor returns :
I would like to know if turboboost can have some limitations in high percentile, because my kpi really worsen when binding new 100%cpu process on idle core.
So even if cpupower report 99.9% of time on c0 at 3.294ghz, can it happens that turboboost in the 0.01% remaining, really lower the power in the core which can impact my application kpi in the high percentile almost x3 my observed latency ?
Is there any papers related to turboboost performance and heavy load ?
Is there any tool better than standard linux tools to monitor turboboost mode on a lower time scale (ie microsecond or nanosecond)?
Thank you for your concern and your replies.
If i did any mistakes i will be committed to correct them.
My mean concern is looking for study, about turboboost limitation.
Let's say i have turboboost activated on my server and that usually on a 16cores cpu (model name : Intel(R) Xeon(R) CPU E5-2690 0 @ 2.90GHz
),
- only 5 or 6 are used 100% with polling process which are binded on distinct cores (one process per core, stuck at 100%cpu (while true do.... style)
- cstate enforced to 0 in bios and linux boot cmdline (intel_idle.max_cstate=0 processor.max_cstate=0 idle=poll)
If i add more processes, in this context cpupower monitor returns :
Code:
|Nehalem || SandyBridge || Mperf
PKG |CORE|CPU | C3 | C6 | PC3 | PC6 || C7 | PC2 | PC7 || C0 | Cx | Freq
0| 0| 0| 0.00| 0.00| 0.00| 0.00|| 0.00| 0.00| 0.00|| 3.28| 96.72| 3292
0| 1| 1| 0.00| 0.00| 0.00| 0.00|| 0.00| 0.00| 0.00|| 1.22| 98.78| 3291
0| 2| 2| 0.00| 0.00| 0.00| 0.00|| 0.00| 0.00| 0.00|| 0.55| 99.45| 3292
0| 3| 3| 0.00| 0.00| 0.00| 0.00|| 0.00| 0.00| 0.00|| 99.99| 0.01| 3292
0| 4| 4| 0.00| 0.00| 0.00| 0.00|| 0.00| 0.00| 0.00|| 99.99| 0.01| 3292
0| 5| 5| 0.00| 0.00| 0.00| 0.00|| 0.00| 0.00| 0.00|| 99.99| 0.01| 3292
0| 6| 6| 0.00| 0.00| 0.00| 0.00|| 0.00| 0.00| 0.00|| 99.99| 0.01| 3292
0| 7| 7| 0.00| 0.00| 0.00| 0.00|| 0.00| 0.00| 0.00|| 99.99| 0.01| 3292
1| 0| 8| 0.00| 0.00| 0.00| 0.00|| 0.00| 0.00| 0.00|| 99.99| 0.01| 3292
1| 1| 9| 0.00| 0.00| 0.00| 0.00|| 0.00| 0.00| 0.00|| 99.99| 0.01| 3292
1| 2| 10| 0.00| 0.00| 0.00| 0.00|| 0.00| 0.00| 0.00|| 99.99| 0.01| 3292
1| 3| 11| 0.00| 0.00| 0.00| 0.00|| 0.00| 0.00| 0.00|| 0.56| 99.44| 3292
1| 4| 12| 0.00| 0.00| 0.00| 0.00|| 0.00| 0.00| 0.00|| 99.99| 0.01| 3292
1| 5| 13| 0.00| 0.00| 0.00| 0.00|| 0.00| 0.00| 0.00|| 99.99| 0.01| 3292
1| 6| 14| 0.00| 0.00| 0.00| 0.00|| 0.00| 0.00| 0.00|| 99.99| 0.01| 3292
1| 7| 15| 0.00| 0.00| 0.00| 0.00|| 0.00| 0.00| 0.00|| 0.36| 99.64| 3293
I would like to know if turboboost can have some limitations in high percentile, because my kpi really worsen when binding new 100%cpu process on idle core.
So even if cpupower report 99.9% of time on c0 at 3.294ghz, can it happens that turboboost in the 0.01% remaining, really lower the power in the core which can impact my application kpi in the high percentile almost x3 my observed latency ?
Is there any papers related to turboboost performance and heavy load ?
Is there any tool better than standard linux tools to monitor turboboost mode on a lower time scale (ie microsecond or nanosecond)?
Thank you for your concern and your replies.
If i did any mistakes i will be committed to correct them.