Introduction

SOFA: Swarm of Functions Analysis Authors: All the contributors of SOFA

Prerequisite

1. Run ./tools/prerequisite.sh to install all the necessary packages and python packages.
2. [OPTIONAL] Run ./tools/empower-tcpdump.sh $(whoami) to make network related events tracable in SOFA. After running this step, it is required to re-login to APPLY THE CHANGES!!!

Verify Permissions

Run the following simple tests to verify the permission settings of executing tcpdump under your permission.

• tcpdump -w sofa.pcap
• tcpdump -r sofa.pcap

Installation

1. Simply run ./install.sh </PATH/TO/INSTALL> to install SOFA on your system. Note that sofa will be appended to the path if the last directory is not sofa.
2. Then, run source </PATH/TO/INSTALL>/sofa/tools/activate.sh to activate SOFA running environment. (Need to be executed on each new shell.)
3. [ALTERNATIVE] Add source </PATH/TO/INSTALL>/sofa/tools/activate.sh in your ~/.bashrc to make this environment available on every shell.

Uninstall

Run bash </PATH/TO/INSTALL>/sofa/tools/uninstall.sh to safely remove/uninstall files of SOFA.

Usages

SOFA supports serveral different usages, like how one can use perf.

Basic Statistics

1. Profile your program by sampling involved CPUs:
   sofa stat "wget http://www.bbc.com/news"
2. Profile your program by sampling all CPUs:
   sofa stat "wget http://www.bbc.com/news" --profile_all_cpus

Performance Visualizations

1. sofa record "wget http://www.bbc.com/news"
2. sofa report
3. Open browser with one of the following links for different visualizations
   • http://localhost:8000
   • http://localhost:8000/cpu-report.html
   • http://localhost:8000/gpu-report.html

Run with different plugins

1. Run sofa record "wget http://www.bbc.com/news" only once to record the events.
2. Run sofa preprocess only once to process the events.
3. Run sofa analyze with A plugin.
4. Run sofa analyze with B plugin.
5. Run sofa analyze with C plugin.

Configurations

SOFA provides options for configurations. Some examples are shown below. Please use sofa --help to see more info.

1. sofa --cpu_filters="idle:black,tensorflow:orange" record "python tf_cnn_benchmarks.py"
2. sofa --gpu_filters="tensorflow:orange" --verbose=True record "python tf_cnn_benchmarks.py"

Examples of Visualization Results:

sofa record "~/cuda_samples/1_Utilities/bandwidthTest/bandwidthTest" sofa record "python tf_cnn_benchmarks.py --num_gpus=8 --batch_size=64 --model=resnet50 --variable_update=parameter_server --num_warmup_batches=1 --num_batches=3" sofa record "python tf_cnn_benchmarks.py --num_gpus=8 --batch_size=64 --model=resnet50 --variable_update=parameter_server --num_warmup_batches=1 --num_batches=3" sofa record "python tf_cnn_benchmarks.py --num_gpus=8 --batch_size=64 --model=alexnet --variable_update=replicated --all_reduce_spec=xring --num_warmup_batches=1 --num_batches=3" sofa record "mpirun -f hosts.txt -n 4 ./compute_pi"