Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly
Sign up
Appearance settings

Commit d7f259b

Browse files
Added code for comparison between x86_64 with arm64 (graviton2) lambda functions
1 parent bb9939c commit d7f259b

File tree

3 files changed

+89
-0
lines changed

3 files changed

+89
-0
lines changed
42.7 KB
Loading[フレーム]

‎xgboost-inference-arm64-docker-lambda/README.md

Lines changed: 15 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -8,6 +8,7 @@ This project contains source code and supporting files for a serverless applicat
88
- train-code - Code for training XGBoost model based on breast cancer dataset.
99
- events - Invocation events that you can use to invoke the function.
1010
- template.yaml - A template that defines the application's AWS resources.
11+
- invoke_x86_64_arm64_lambdas.py - Code to invoke x86_64 and Arm64 (Graviton2) Lambda Functions and compare performance.
1112

1213
The application uses several AWS resources, including Lambda functions. These resources are defined in the `template.yaml` file in this project. You can update the template to add AWS resources through the same deployment process that updates your application code.
1314

@@ -63,6 +64,20 @@ xgboost-inference-arm64-docker-lambda$ sam local invoke XGBoostInferenceArm64Fun
6364

6465
![arm64 Architecture](../img/xgboost_arm_64_arch_view.png)
6566

67+
## Compare performance of x86_64 Lambda Function with Arm64 (Graviton2) one
68+
69+
You can achieve up to 34% better price/performance with AWS Lambda Functions powered by AWS Graviton2 processor.
70+
71+
To test the performance improvements of the XGBoost x86_64 Lambda Function with an Arm64 (Graviton2) one yourself, please execute the following steps:
72+
1. Deploy the XGBoost x86_64 Lambda Function, as described [here](../xgboost-inference-docker-lambda/)
73+
2. Open `invoke_x86_64_arm64_lambdas.py` file.
74+
3. Replace `<XGBoost_x86_64_Lambda_ARN>` with XGBoost x86_64 Lambda Function ARN.
75+
4. Replace `<XGBoost_arm64_Lambda_ARN>` with XGBoost Arm64 (Graviton2) Lambda Function ARN.
76+
5. Run the following command: `python invoke_x86_64_arm64_lambdas.py`
77+
78+
![xgboost x86_64 vs. arm64 improvement](../img/xgboost_x86_64_arm64_improvement.png)
79+
80+
**We can see that the Arm64 (Graviton2) Lambda Function has performance improvements of 36% over the x86_64 one!**
6681

6782
## Testing your Lambda function in the Cloud
6883

Lines changed: 74 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,74 @@
1+
import boto3
2+
import json
3+
import base64
4+
import re
5+
import sys
6+
7+
client = boto3.client('lambda')
8+
9+
f= open('./events/event.json')
10+
json_input = json.load(f)
11+
12+
13+
def call_XGBoost_x86_64_Lambda():
14+
response = client.invoke(
15+
FunctionName='<XGBoost_x86_64_Lambda_ARN>',
16+
InvocationType='RequestResponse',
17+
LogType='Tail',
18+
Payload=json.dumps(json_input)
19+
)
20+
return(find_duration(response))
21+
22+
23+
def call_XGBoost_arm64_Lambda():
24+
response = client.invoke(
25+
FunctionName='<XGBoost_arm64_Lambda_ARN>',
26+
InvocationType='RequestResponse',
27+
LogType='Tail',
28+
Payload=json.dumps(json_input)
29+
)
30+
31+
return(find_duration(response))
32+
33+
34+
def find_duration(response):
35+
log_result = base64.b64decode(response['LogResult']).decode('utf-8')
36+
result = re.findall(r'Duration: (\d+\.\d+) ms', log_result)
37+
return (float(result[0]))
38+
39+
# Warm the two Lambda Functions first
40+
print("Warming x86_64 and arm64 Lambda Functions")
41+
call_XGBoost_x86_64_Lambda()
42+
call_XGBoost_arm64_Lambda()
43+
44+
print("Warming x86_64 and arm64 Lambda Functions - Done")
45+
print("Sending events to x86_64 and arm64 Lambda Functions")
46+
47+
counter=1
48+
num_calls=100
49+
total_duration_x86_64=0
50+
total_duration_arm64=0
51+
52+
while counter < num_calls:
53+
duration_x86_64 = call_XGBoost_x86_64_Lambda()
54+
# print(f'call_XGBoost_x86_64_Lambda duration: {duration_x86_64}')
55+
total_duration_x86_64 = total_duration_x86_64 + duration_x86_64
56+
57+
duration_arm64 = call_XGBoost_arm64_Lambda()
58+
# print(f'call_XGBoost_arm64_Lambda duration: {duration_arm64}')
59+
total_duration_arm64 = total_duration_arm64 + duration_arm64
60+
61+
sys.stdout.write(('=' * counter) + ('' * (num_calls - counter)) + ("\r [ %d" % counter + "% ] "))
62+
sys.stdout.flush()
63+
64+
counter = counter + 1
65+
66+
print("\nSending events to x86_64 and arm64 Lambda Functions - Done")
67+
68+
avg_duration_x86_64 = total_duration_x86_64/num_calls
69+
avg_duration_arm64 = total_duration_arm64/num_calls
70+
improvement_percentage= "{:.0%}".format(1 - (avg_duration_arm64 / avg_duration_x86_64))
71+
72+
print('Average duration x86_64: {:.2f} ms'.format(total_duration_x86_64/num_calls))
73+
print('Average duration arm64: {:.2f} ms'.format(total_duration_arm64/num_calls))
74+
print(f'*** Improvement of arm64 (Graviton2) over x86_64: {improvement_percentage} ***')

0 commit comments

Comments
(0)

AltStyle によって変換されたページ (->オリジナル) /