-
Notifications
You must be signed in to change notification settings - Fork 28
Running Experiments on a Cloud Instance
Running a batch experiment may take few minutes to hours, so we run it on a cloud instance to avoid running such computational heavy job on our local machine for a long period.
We skipped the details of allocating a Cloud Instance; instead, we start from a running instance and a given key to access it.
$ chmod 600 <folder_path_to_where_your_key_stores>/asset2.pem
$ ssh -i <folder_path_to_where_your_key_stores>/asset2.pem [email protected]
To avoid our session be lost after we disconnect, we always create a new session or resume the previous one before starting.
$ screen -RR
$ cd airport-simulation
At the time this document is composed, we have 4 batch experiments.
-
batch_plans/sfo-terminal-2-uc-failed-spot.yaml
: Uncertainty experiment on counting the failure rate -
batch_plans/sfo-terminal-2-uc-success-spot.yaml
: Uncertainty experiment on generating the output metrics -
batch_plans/sfo-terminal-2-rt-failed.yaml
: Reschedule exerpiment on counting the failure rate -
batch_plans/sfo-terminal-2-rt-success.yaml
: Reschedule exerpiment on generating the output metrics
You can edit them using vim
.
Once we've edited the batch plan, we can run it now! For example:
$ ./simulator.py -f batch_plans/sfo-terminal-2-uc-failed-spot.yaml
Starting the simulation in batch mode
Running simulation with uncertainty.prob_hold = 0.000000 (nth = 0)
2018-05-02 18:49:12,849 Done
...
Batch run results are stored in the batch_output/
folder. You can able to find a subfolder that is as same as your batch plan name.
For multiple individual simulations executed by a batch run, you can find the results under output/
To download the data to a local machine, please use scp
like the following command on your local machine, or a visualized tool like CyberDuck.
$ scp -i <folder_path_to_where_your_key_stores>/asset2.pem -r [email protected]:~/airport-simulation/batch_output/<batch_output_name>/ .
Step 1: Run the following command on the cloud instance:
./visualization/server.py
Step 2: Open the browser on your local machine and visit: http://35.170.55.208:5000/
Step 3: Select an experiment result from the left bottom dropdown list. It should start to load the data. Please wait for around 1~2 minutes for it to load the data. After the data is loaded, the right bottom should display "00:00" instead of "loading". Then, you can click on the arrow buttons or "auto-play" button to view the results.