This is the Synthetic function generator from the manuscript "Sizeless: Predicting the optimal size of serverless functions". It generates AWS Lambda deployable functions by randomly combining commonly occurring function segments. The generated functions are instrumented with a monitoring functionality. Besides the generation it also provides the possibility to directly benchmark the resulting functions. Overall, it allows the generation of a large number of serverless functions and benchmarking them.
This tool is implemented as a Command Line Interface (CLI) which allows automating experiments, e.g. using the CLI in scripts.
The Synthetic Function Generator makes its first appearance in the manuscript "Sizeless: Predicting the optimal size of serverless functions", which can currently be found under https://arxiv.org/abs/2010.15162. If you use the Synthetic Function Generator please cite the following publication:
@misc{eismann2020sizeless,
title={Sizeless: Predicting the optimal size of serverless functions},
author={Simon Eismann and Long Bui and Johannes Grohmann and Cristina L. Abad and Nikolas Herbst and Samuel Kounev},
year={2020},
eprint={2010.15162},
archivePrefix={arXiv},
primaryClass={cs.DC}
}
Requirements:
- golang installed (tested for versions 1.14+)
Inside the function-generator directory run the command go build ..
By default this results in a file named synthetic-function-generator.
When using the CLI only to generate functions, you will only need the steps 4 and 5 in order to make the AWS Lambda functions deployable. To be able to use the load running functionality, however, there is some setup involved. This is due to some of the AWS tools being only available as CLI and due to the setup and teardown scripts being NodeJS scripts.
You will need a working NodeJS runtime.
This means that you will have to setup your environment so that it is possible to run the node command.
It is generally recommended to use the LTS version.
With node -v you can check both whether you have a working NodeJS runtime and which version you have.
The load runner makes use of the AWS CLI to cleanup the generated AWS Lambda functions. See the official documentation on how to install the AWS CLI.
The Serverless Application Model CLI needs to be installed and invokable. See the official AWS SAM CLI documentation on how to do so. This is needed to deploy the generated AWS Lambda functions.
Both the AWS CLI and SAM CLI read the AWS Credentials that are needed to perform actions on AWS from either environment variables or by convention from a file in $HOME/.aws/credentials.
Make sure that a valid Access Key and Secret Key can be found by both CLIs.
See the official documentation for more details on configuring the AWS CLI.
The monitoring add-on has some dependencies on certain npm packages.
To avoid that each synthetic function package has to bring these dependencies, the use of a dependency layer proved to be the most efficient way.
The file dependencyLayer.zip represents a ready-to-deploy package to add the required dependencies as a Lambda Layer. Follow the instructions on AWS Console -> Lambda -> Additional resources -> Layers to deploy the package.
The dependency layer will be assigned a Version ARN, which is needed to generate functions using the CLI.
For more information about AWS Lambda Layers, see here.
The synthetic Lambda functions need to be assigned a role ARN to allow the function to access other AWS services. Either create a new role that defines what services the Lambda functions are allowed to access or use an existing role. Make sure that AWS Lambda is listed in the Trusted Entities section so that it can be used by Lambda. To generate synthetic functions using the CLI the Role ARN will be needed.
The CLI offers descriptions and help texts for each command as well as an overall usage description:
CLI for generating AWS Lambda deployable synthetic function packages
This application can be used to generate a large amount of synthetic functions
that can be deployed to AWS Lambda for all kinds of measurements.
It also offers the capability to run invocations against those deployed functions.
Usage:
synthetic-function-generator [command]
Available Commands:
clean Command to clean the build directory
generate Used to create serverless function packages
help Help about any command
runload Start load run for measurement data
Flags:
-h, --help help for synthetic-function-generator
Use "synthetic-function-generator [command] --help" for more information about a command.
This command generates AWS Lambda deployable serverless function artifacts.
The generated artifacts are compatible for use with the runload command.
Usage:
synthetic-function-generator generate [flags]
Flags:
-d, --dependency-layern-arn string The ARN of the dependency layer, see README (required)
--exclude string Path to file containing roll strings to be excluded from generation (use --save flag for an example)
-f, --func-segments string Path to function segments to be used for generation (required)
-h, --help help for generate
-l, --lambda-role-arn string The ARN of the lambda role, see README (required)
-m, --max-roll int Maximum number of rolled function segments (ignored if replayFile is provided) (default 3)
-n, --num-funcs int Number of functions to generate (ignored if replayFile is provided) (default 1)
--replay string Path to file containing roll strings to be regenerated (use --save flag for an example)
--save Whether the generated function combinations should be saved
-s, --sizes ints Specify function sizes to be generated, need to be supported by the platform! (default [128,416,704,992,1280,1568,1856,2144,2432,2720,3008])
This command drives load on the generated functions on lambda.
It writes the results on the local filesystem specified by the flag --result-dir for later evaluation.
Usage:
synthetic-function-generator runload [flags]
Flags:
-d, --duration int Target duration in seconds (default 10)
-f, --func-dir string Directory containing the lambda functions (required)
-h, --help help for runload
-p, --req-per-sec int Target requests per second (default 50)
-r, --result-dir string Directory for the measurement results (default "./result-data")
-w, --workers int Number of parallel workers (default 5)
This command can be used to delete all generated functions
Usage:
synthetic-function-generator clean [flags]
Flags:
-h, --help help for clean