Quick Start

TaPS is a standardized framework for evaluating task-based execution frameworks and data management systems using a suite a real and synthetic scientific applications.

Installation

git clone https://github.com/proxystore/taps
cd taps
python -m venv venv
. venv/bin/activate
pip install -e .

Documentation on installing for local development is provided in Contributing.

Usage

Applications can be executed from the CLI.

python -m taps.run --app {name} {args}

See python -m taps.run --help for a list of applications.

Example App

The Cholesky Factorization app, for example, can be run using --app cholesky and the required arguments.

python -m taps.run --app cholesky --app.matrix_size 100 --app.block_size 25

Note

This cholesky app requires having installed TaPS with the [cholesky] option.

pip install -e .[cholesky]

Many execution options can be altered directly from the command line. The above example, by default, used a ProcessPoolExecutor but we can switch to a different executor easily with the --engine.executor flag.

python -m taps.run \
    --app cholesky --app.matrix_size 100 --app.block_size 25 \
    --engine.executor thread-pool

Config Files

Alternatively, applications can be configured using a TOML configuration file.

[app]
name = "cholesky"
matrix_size = 100
block_size = 25

[engine]
task_record_file_name = "tasks.jsonl"

[engine.executor]
name = "process-pool"
max_processes = 10

[run]
dir_format = "runs/{name}_{executor}_{timestamp}"

[logging]
file_level = "WARNING"
file_name = "log.txt"
level = "INFO"

To execute from a config, use the -c/--config option.

python -m taps.run --config config.toml

Options provided via the CLI will override those options present in a config file.

Apps

Checkout the Application Guides to learn about all of the different benchmarking applications provided by TaPS.