COSMOS - Framework for Combining Simulation and Optimization Software
It has three main functions:
- optimize application input and parameters
- provide an optimal scheduling policy for all sub-tasks of the application
- analyze the application results.
Requirements:
- python2.7
Steps for using COSMOS with broker running on localhost:
-
go to task_examples/pi_computing and compile a dummy program that computes pi in N iterations
-
edit the configs/broker-config.json file and add:
- at least one host
- choose scheduling method (at this moment, only min-min is implemented)
- set rootfs field to the folder where you have your application / scripts / input (in our case is /path/to/repo/task_examples/pi_computing)
-
[Optional step] edit configs/optimizer-config.json and add optimization methods for each task. You can have one pre optimization method that gets executed before the task is scheduled and one post method that is executed after the task finishes execution.
-
edit configs/tasks.json (in our case pi_tasks.json) and for each task specify:
- command: this is what gets executed in the background thread
- args: command arguments used when starting execution
- priority: used when using the priority-scheduling method from broker config (not implemented yet)
- length: estimated task length used when computing the load for selecting the host
-
./main.py --commands-file configs/pi_tasks.json
If you have any suggestions or ideas please feel free to contribute either by pull requests or sending patch on email.