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.

Sign up for GitHub

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

add relative standard deviation to aggregated test execution metrics #681

Merged
merged 4 commits into from
Oct 25, 2024
Merged

add relative standard deviation to aggregated test execution metrics #681

Changes from 1 commit
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
4f06e0f
add relative standard deviation to aggregated test execution metrics
OVI3D0 Oct 14, 2024
0467ae8
address govind's comments
OVI3D0 Oct 22, 2024
4912799
update error message for rsd calculation
OVI3D0 Oct 23, 2024
c0a25c5
add type hint for values in calculate_rsd args
OVI3D0 Oct 24, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
38 changes: 26 additions & 12 deletions osbenchmark/aggregator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,5 @@
import os
import statistics
from typing import Any, Dict, List, Union
import uuid

Expand Down Expand Up @@ -123,14 +124,17 @@ def build_aggregated_results(self):
op_metric = {
"task": task,
"operation": task,
"throughput": aggregated_task_metrics["throughput"],
"latency": aggregated_task_metrics["latency"],
"service_time": aggregated_task_metrics["service_time"],
"client_processing_time": aggregated_task_metrics["client_processing_time"],
"processing_time": aggregated_task_metrics["processing_time"],
"error_rate": aggregated_task_metrics["error_rate"],
"duration": aggregated_task_metrics["duration"]
}
for metric in self.metrics:
op_metric[metric] = aggregated_task_metrics[metric]
if isinstance(aggregated_task_metrics[metric], dict):
mean_values = [v['mean'] for v in task_metrics[metric]]
rsd = self.calculate_rsd(mean_values)
op_metric[metric]['mean_rsd'] = rsd
else:
rsd = self.calculate_rsd(task_metrics[metric])
op_metric[f"{metric}_rsd"] = rsd

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Some comments will be helpful here -- the first clause is for the standard metrics, the second for derived ones like error rate and duration, etc.

The documentation should indicate the rationale as to why RSD is computed for the mean, instead of the median.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'll also speak to nate about these changes👍

aggregated_results["op_metrics"].append(op_metric)

# extract the necessary data from the first test execution, since the configurations should be identical for all test executions
Expand Down Expand Up @@ -184,8 +188,8 @@ def calculate_weighted_average(self, task_metrics: Dict[str, List[Any]], iterati
weighted_metrics = {}

for metric, values in task_metrics.items():
weighted_metrics[metric] = {}
if isinstance(values[0], dict):
weighted_metrics[metric] = {}
for item_key in values[0].keys():
if item_key == 'unit':
weighted_metrics[metric][item_key] = values[0][item_key]
Expand All @@ -194,18 +198,28 @@ def calculate_weighted_average(self, task_metrics: Dict[str, List[Any]], iterati
if iterations > 1:
weighted_sum = sum(value * iterations for value in item_values)
total_iterations = iterations * len(values)
weighted_metrics[metric][item_key] = weighted_sum / total_iterations
weighted_avg = weighted_sum / total_iterations
else:
weighted_metrics[metric][item_key] = sum(item_values) / len(item_values)
weighted_avg = sum(item_values) / len(item_values)
weighted_metrics[metric][item_key] = weighted_avg
else:
if iterations > 1:
weighted_sum = sum(value * iterations for value in values)
total_iterations = iterations * len(values)
weighted_metrics[metric] = weighted_sum / total_iterations
weighted_avg = weighted_sum / total_iterations
else:
weighted_metrics[metric] = sum(values) / len(values)
weighted_avg = sum(values) / len(values)
weighted_metrics[metric] = weighted_avg
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why the special case for iterations == 1? Doesn't this reduce to the same computation? And if the iteration number is the same for every element in values, the special-casing might not be needed.

Copy link
Member Author

@OVI3D0 OVI3D0 Oct 22, 2024 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The special case is fixed, but I want to include support for custom iteration values in the future so I think that should stay for now.


return weighted_metrics

def calculate_rsd(self, values):
if not values:
return 0
OVI3D0 marked this conversation as resolved.
Show resolved Hide resolved
mean = statistics.mean(values)
std_dev = statistics.stdev(values) if len(values) > 1 else 0
OVI3D0 marked this conversation as resolved.
Show resolved Hide resolved
return (std_dev / mean) * 100 if mean != 0 else 0
OVI3D0 marked this conversation as resolved.
Show resolved Hide resolved

def test_execution_compatibility_check(self) -> None:
first_test_execution = self.test_store.find_by_test_execution_id(list(self.test_executions.keys())[0])
workload = first_test_execution.workload
Expand Down
Loading