Skip to content

Latest commit

 

History

History
262 lines (209 loc) · 10.2 KB

README.md

File metadata and controls

262 lines (209 loc) · 10.2 KB

Start-up Components Experiments

The Function Start-up Components experiments points to all conducted experiments to generate the results for the paper section 4.2.1.

These experiments evaluate the NOOP, Image-Resizer and Markdown functions using the Vanilla and Prebaking techniques, but measuring the following start-up phases:

  1. CLONE;
  2. EXEC;
  3. RTS;
  4. APPINIT.

Setup

Before executing the experiments, its required to declare two environment variables:

  1. CRIU_BINARY_PATH - the absolute path to CRIU's executable file.
  2. JAVA_BINARY_PATH - the absolute path to Java 8 executable file.

First, you need to find the absolute path to CRIU and Java 8 executable files. The whereis command can report it to you. The following snippet shows whereis execution examples for criu and java.

$ whereis java
java: /usr/bin/java /usr/share/java /usr/share/man/man1/java.1.gz
$ whereis criu
criu: /usr/local/sbin/criu

Note that for criu, the whereis command found only the path /usr/local/sbin/criu. However, whereis found multiple paths for java. In this case, you can assume that the first one is the correct.

Then, after finding the absolute path to CRIU and Java 8 binaries, you can declare the environment variables:

# CRIU_BINARY_PATH=/usr/local/sbin/criu
# JAVA_BINARY_PATH=/usr/bin/java

Please note that, all the following bash commands may require super user rights.

NOOP

The NOOP is a do-nothing function. For the NOOP function we must perform two experiments: Vanilla and Prebaking.

Inside the prebaking-middleware-2020 directory, create a Load Generator Config file named noop-config.json which holds the following content:

{
    "Requests": 200,
    "RequestSpec": {
        "Method": "GET",
        "Path": "/"
    }
}

NOOP Vanilla

After the configuration steps, you can execute the Start-up Components NOOP Vanilla experiment using the following commands:

# cd serverless-handlers
# bash run-experiment.sh java noop 200 no-criu ../noop-config.json \
    --executor_process_name=${JAVA_BINARY_PATH} > log.out 2> log.err

The files log.out and log.err file will contain the stdout and stderr execution logs. You can consult these files to identify any unexpected error during the experiment execution.

If the above commands execute successfully, then an output CSV file will be created containing all the collected metrics during the experiment execution. A successful output file will hold the metrics data for 200 executions with 200 requests each.

Please quick check this file and rename it to startup-components-noop-vanilla.csv.

NOOP Prebaking

Now, you can execute the Start-up Components NOOP Prebaking experiment using the following commands:

# cd serverless-handlers
# bash run-experiment.sh java noop 200 criu ../noop-config.json \
    --executor_process_name=${CRIU_BINARY_PATH} > log.out 2> log.err

The files log.out and log.err file will contain the stdout and stderr execution logs.

If the above commands execute successfully, then an output CSV file will be created containing all the collected metrics during the experiment execution. A successful output file will hold the metrics data for 200 executions with 200 requests each.

Please quick check the output file and rename it to startup-components-noop-prebaking.csv.

Image-Resizer

The Image-Resizer app scales down an image, then we need to download the image used to evaluate the Image-Resizer application. The following command describes how to do it.

$ wget https://i.imgur.com/BhlDUOR.jpg

For the Image-Resizer function we must perform two experiments: Vanilla and Prebaking.

Inside the prebaking-middleware-2020, create a Load Generator Config file named image_resizer-config.json, holding the below content. The EnvVars property must declare the variables scale and image_path. The image_path var should point to the absolute path of the downloaded image named BhlDUOR.jpg.

{
    "EnvVars": ["scale=0.1", "image_path={PathFromRoot}/prebaking-middleware-2020/BhlDUOR.jpg"],
    "Requests": 200,
    "RequestSpec": {
        "Method": "GET",
        "Path": "/"
    }
}

Image-Resizer Vanilla

After the configuration steps, you can execute the Start-up Components Image-Resizer Vanilla experiment using the following commands:

# cd serverless-handlers
# bash run-experiment.sh java thumbnailator 200 no-criu ../image_resizer-config.json \
    --executor_process_name=${JAVA_BINARY_PATH} > log.out 2> log.err

The files log.out and log.err file will contain the stdout and stderr execution logs.

If the above commands execute successfully, then an output CSV file will be created containing all collected metrics. A successful output file will hold the metrics data for 200 executions with 200 requests each.

Please quick check this file and rename it to startup-components-image_resizer-vanilla.csv.

Image-Resizer Prebaking

Now, you can execute the Start-up Components Image-Resizer Prebaking experiment using the following commands:

# cd serverless-handlers
# bash run-experiment.sh java thumbnailator 200 criu ../image_resizer-config.json \
    --executor_process_name=${CRIU_BINARY_PATH} > log.out 2> log.err

The files log.out and log.err file will contain the stdout and stderr execution logs.

If the above commands execute successfully, then an output CSV file will be created containing all collected metrics. A successful output file will hold the metrics data for 200 executions with 200 requests each.

Please quick check this file and rename it to startup-components-image_resizer-prebaking.csv.

Markdown

The Markdown function renders markdown files into HTML. To evaluate this function we used the OpenPiton README.md.

The following command downloads the file and rename it to OpenPiton-README.md:

$ wget -O OpenPiton-README.md https://raw.githubusercontent.com/PrincetonUniversity/openpiton/openpiton/README.md

For the Markdown function we also should perform two experiments: Vanilla and Prebaking.

Inside the prebaking-middleware-2020 directory, you need to create a Load Generator Config file named markdown-config.json, holding the below content. The Bodyfilepath field from the RequestSpec property must point to the absolute path of the OpenPiton README.md.

{
    "Requests": 200,
    "RequestSpec": {
        "Method": "POST",
        "Path": "/",
        "Bodyfilepath": "{PathFromRoot}/prebaking-middleware-2020/OpenPiton-README.md"
    }
}

Markdown Vanilla

After the configuration steps, you can execute the Start-up Components Markdown Vanilla experiment using the following commands:

# cd serverless-handlers
# bash run-experiment.sh java markdown 200 no-criu ../markdown-config.json \
    --executor_process_name=${JAVA_BINARY_PATH} > log.out 2> log.err

The files log.out and log.err file will contain the stdout and stderr execution logs.

If the above commands execute successfully, then an output CSV file will be created containing all collected metrics. A successful output file will hold the metrics data for 200 executions with 200 requests each.

Please quick check this file and rename it to startup-components-markdown-vanilla.csv.

Markdown Prebaking

Finally, you can execute the Start-up Components Markdown Prebaking experiment using the following commands:

# cd serverless-handlers
# bash run-experiment.sh java markdown 200 criu ../markdown-config.json \
    --executor_process_name=${CRIU_BINARY_PATH} > log.out 2> log.err

The files log.out and log.err file will contain the stdout and stderr execution logs.

If the above commands execute successfully, then an output CSV file will be created containing all collected metrics. A successful output file will hold the metrics data for 200 executions with 200 requests each.

Please quick check this file and rename it to startup-components-markdown-prebaking.csv.

Data Analysis

After executing all the experimental scenarios, to analyze the data will be required to merge all the results artifacts into a file named java_treated_startup_bpftrace.csv, the following commands automate this task for you:

cp ../startup-components/merge.R .
Rscript merge.R
mkdir data
mv java_treated_startup_bpftrace.csv data/

The documentation of the file java_treated_startup_bpftrace.csv can be accessed here.

After it, you can copy and execute the analysis script by running the following commands:

cp ../eval/startup-components/analysis.R .
Rscript analysis.R

The analysis script will load the file java_treated_startup_bpftrace.csv and generate the paper artifacts described here.