Automatic DAG Generator is customized for Jupiter Orchestrator (available here: [https://github.com/ANRGUSC/Jupiter]).
rand_task_gen.py
will work with task_config.yml
to generate a dummy application used for Jupiter.
There are three main concepts in this file: depth
, width_min, width_max
and total_num
.
depth
parameter is used for define how many levels of the DAG, which not include the very beginning task and the last task. In other words,
If you want to create a total 6 level DAG, you should put depth: 4
.
width_min
and width_max
are used for define how many tasks in each level except the beginning and the last tasks. The number of tasks will between those two
parametes per level.
total_num
is used for specify a total number of tasks in your DAG. For example, if you want to generate a total 10 tasks in the DAG including the beginning and the last tasks,
you should put total_num: 8
, which 8 comes from (10 - 2). If there is a conflict between total_num
and other parametes, the DAG will be generated with the first priority of total_num
.
To run this script, you should set --conf task_config.yml
as the parameter. It will generate the files below:
- dummy_app/scripts -- scripts used in app_specific_files
- config.json -- used along with dummy_app/scipts
- configuration.txt -- used in CIRCE
- communication.txt -- NOT used for Jupiter, but for users to track tasks relations
- dag.png
You should also create a sample_input/
folder and have some input files there. More details please follow the instructions on how to use Jupiter at [http://jupiter.readthedocs.io/].
This code uses the output DAG of rand_task_gen.py
(by Diyi), which is used to generate a random DAG, to generate the corresponding dummy application for Jupiter. To run this script, you should set --conf task_config.yml
as the parameter. It will generate the dummy_app
folder and all required content which can be used as a sample application for Jupiter. The dummy_app
has been tested and work with Jupiter Version 3 & 4.
This material is based upon work supported by Defense Advanced Research Projects Agency (DARPA) under Contract No. HR001117C0053. Any views, opinions, and/or findings expressed are those of the author(s) and should not be interpreted as representing the official views or policies of the Department of Defense or the U.S. Government.