The user can modify NG_CPUS_PER_TASK #1336
Open
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The environment variable NG_CPUS_PER_TASK modifies the number of CPUs to be used per task.
facebook-github-bot
added
the
CLA Signed
jrapin
reviewed
Jan 19, 2022
nevergrad/benchmark/xpbase.py
Outdated
| @@ -57,7 +59,11 @@ def __init__( | |||
| self.optimizer = optimizer | |||
| self.budget = budget | |||
| self.num_workers = num_workers | |||
| self.executor = execution.MockedTimedExecutor(batch_mode) | |||
| ng_cpus_per_task = int(os.getenv("NG_CPUS_PER_TASK", "1")) | |||
| if ng_cpus_per_task > 0: | |||
There was a problem hiding this comment.
Suggested change
| if ng_cpus_per_task > 0: | |
| if ng_cpus_per_task > 1: |
jrapin
reviewed
Jan 26, 2022
| @@ -57,15 +59,20 @@ def __init__( | |||
| self.optimizer = optimizer | |||
| self.budget = budget | |||
| self.num_workers = num_workers | |||
| self.executor = execution.MockedTimedExecutor(batch_mode) | |||
| ng_cpus_per_task = int(os.getenv("NG_CPUS_PER_TASK", "1")) | |||
There was a problem hiding this comment.
This is only for benchmark so let's make it clear
Suggested change
| ng_cpus_per_task = int(os.getenv("NG_CPUS_PER_TASK", "1")) | |
| ng_cpus_per_task = int(os.getenv("NG_BENCHMARK_CPUS_PER_TASK", "1")) |
| self.executor = execution.MockedTimedExecutor(batch_mode) | ||
| ng_cpus_per_task = int(os.getenv("NG_CPUS_PER_TASK", "1")) | ||
| if ng_cpus_per_task > 1: | ||
| self.executor: tp.Any = futures.ThreadPoolExecutor(max_workers=ng_cpus_per_task) |
There was a problem hiding this comment.
I'm still very much against this since it
- prevents reproducibility
- adds complexity which may create a lot of issues (eg: not sure that it will checkpoint correctly if preempted)
- assumes the benchmarks are threadsafe, which is not always true on our side and possibly on the dependency side
Can you quantify the speed gain?
Python executes one thread at a time for most operations, so it may just waste time for creating threads without going faster (depending on what is happening in the underlying code)
Moving to a ProcessPoolExecutor could help for parallelization, but I would expect a big overhead so that it could be unhelpful most of the time
jrapin
reviewed
Jan 26, 2022
|
|
||
| @property | ||
| def name(self) -> str: | ||
| return self.optimizer if isinstance(self.optimizer, str) else repr(self.optimizer) | ||
|
|
||
| @property | ||
| def batch_mode(self) -> bool: | ||
| return self.executor.batch_mode | ||
| return hasattr(self.executor, "batch_mode") and self.executor.batch_mode |
There was a problem hiding this comment.
Suggested change
| return hasattr(self.executor, "batch_mode") and self.executor.batch_mode | |
| return isinstance(self.executor, execution.MockedTimedExecutor) and self.executor.batch_mode |
There was a problem hiding this comment.
checking hasattr is not safe, if we update the code that could break without notice, it's much safer to test against a class
| @@ -207,7 +214,9 @@ def run(self) -> tp.Dict[str, tp.Any]: | |||
| def _log_results(self, pfunc: fbase.ExperimentFunction, t0: float, num_calls: int) -> None: | |||
| """Internal method for logging results before handling the error""" | |||
| self.result["elapsed_time"] = time.time() - t0 | |||
| self.result["pseudotime"] = self.optimsettings.executor.time | |||
| self.result["pseudotime"] = ( | |||
| self.optimsettings.executor.time if hasattr(self.optimsettings.executor, "time") else float("nan") | |||
There was a problem hiding this comment.
Suggested change
| self.optimsettings.executor.time if hasattr(self.optimsettings.executor, "time") else float("nan") | |
| self.optimsettings.executor.time if isinstance(self.executor, execution.MockedTimedExecutor) else float("nan") |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The environment variable NG_CPUS_PER_TASK modifies the number of CPUs to be used per task.
Types of changes
Motivation and Context / Related issue
Possibility to parallelize inside each run in benchmarks.
How Has This Been Tested (if it applies)
Checklist