scripts

To run the benchmarks:

• Make sure nix is installed. I have instead placed the nix-portable executable in the home directory. If you have nix installes natively, you can omit all references to the binary and directly start the commnd with nix ...
• I have cloned this repository to ~/dfd.

Pre-Setup

Gather most required files by running

nix build gitlab:torbeneims/ddfd-evaluation#all

on a nix-enabled system (this does not work with nix-portalbe). Then copy the files from the result link to the scripts folder in this repository. Make sure not to overwrite files.

Then run these commands:

Setup

# Enter dev shell with required python packages
cd ~/dfd/scripts/data; ../../../nix-portable nix develop ../..#pyShell # <-- The # before is not a comment
sh ../../make_datasets.sh

# Exit dev shell
^d # or exit

In different shells, start the spark master and workers:

screen -RR -D sparkm
cd ~/dfd; ../nix-portable nix develop .#default -c master & > spark_master.log

screen -RR -D sparkc
cd ~/dfd; sh 8_spark_workers.sh > spark_clients.log

Make sure the workers are only started when necessary, it makes sense to execute the commands in evaluate.sh individually or comment them out

Evaluation

# Start a screen
screen -RR -D hyperfine
# Enter a different dev shell for evaluation
cd ~/dfd/scripts; ../../nix-portable nix develop ..#default
sh ../evaluate.sh
^d 

Result collection

cd ~/dfd/scripts; ../../nix-portable nix develop ..#pyShell
sh ../collect_results.sh