Add note about --cpu-list in README #333
Conversation
README.md
Outdated
| assumed that your machine has at least 5 CPU cores. You may want to adjust the | ||
| `--cpu-list` parameter based on whatever number of CPU cores your machine has | ||
| or if you don't know how many cores it has then consider setting the parameter | ||
| to `--cpu-list 1` to use only one CPU core. |
There was a problem hiding this comment.
--cpu-list 5 actually runs the benchmark on only one core, which is core number 5. Making it --cpu-list 1 would run it on core number 1 instead. This line sort of makes it sound like the former case runs the benchmarks on more than one core, which it doesn't. Would be nice to clarify that here.
There was a problem hiding this comment.
Your feedback is much appreciated. I have made the some edits that take your points in consideration feel free to take a look at them whenever you can (:
moazzammoriani
force-pushed
the
Add_note_on_cpu-list_in_README
branch
5 times, most recently
from
274f440
to
52372fa
Compare
|
Seems like the loadavg wouldn't go down so the CI build was killed |
moazzammoriani
force-pushed
the
Add_note_on_cpu-list_in_README
branch
from
52372fa
to
c0d1036
Compare
| @@ -212,6 +212,18 @@ The scripts for latencies are present in the `pausetimes/` directory. The | |||
| `pausetimes_trunk` Bash script obtains the latencies for stock OCaml and the | |||
| `pausetimes_multicore` Bash script for Multicore OCaml. | |||
|
|
|||
| **NOTE**: You may need to modify the `--cpu-list` parameter in the config file | |||
There was a problem hiding this comment.
It is useful to add a note about --cpu-list to the README, but, this is not sufficient. One needs to also disable hyperthreading, setup isolcpus, remove RCU callbacks etc. as mentioned in the Notes on hardware and OS settings for Linux benchmarking section at https://github.com/ocaml-bench/ocaml_bench_scripts.
| assumed that your machine has a fifth core to use for the build. If your build | ||
| fails then a possible reason might be that your machine doesn't have a fifth | ||
| CPU core. In such a situation you might want to modify the parameter to use | ||
| any number of cores less than or equal to the number of cores your machine |
There was a problem hiding this comment.
One must use only the isolated CPUs.
| any number of cores less than or equal to the number of cores your machine | ||
| has. If you do not know the number of cores your machine has then your safest | ||
| option is to modify to the parameter to `--cpu-list 1` to simply use the first | ||
| core. |
There was a problem hiding this comment.
One needs to know the machine on which the cores are running. Usually, the first few cores are non-isolated CPUs, and are used by the OS.
There was a problem hiding this comment.
This reasoning stems from the fact that it was usually recommended to set --cpu-list to 1 by you or @Sudha247. Inference filled the rest of the gaps I suppose. Would you say that I should remove this suggestion of setting it to 1 altogether and instead suggest to set --cpu-list to be some non-isolated CPU?
There was a problem hiding this comment.
Just setting the --cpu-list is not sufficient to run the parallel benchmarks. A detailed information on the setup is required, and will be added as part of HACKING.md.
There was a problem hiding this comment.
Then perhaps this note should either point to HACKING.md or should be removed altogether I suppose.
There was a problem hiding this comment.
A note on --cpu-list will be added to HACKING.md.
Fixes #332.