Quanterference

Generating evaluation results:

• generate_eval_results.py is a runnable script which does the following:
  • Trains/tests a model from scratch for each benchmark workload dataset resulting in recreation of the confusion matrices shown in Figure 3.
  • Trains/tests a model from scratch for each real application workload dataset resulting in recreation of the confusion matrices shown in Figure 5.
  • Trains/tests a 3-bin classification model on the io500 benchmark dataset resulting in the recreation of the confusion matrix shown in Figure 4.
  • Processes the enzo analysis trace data (from section II) to recreate plots a and b from Figure 1.

Note: Each of the generated figures will be saved to the eval_results directory inluding the Figure which it represents in brackets. ie. '...[Fig5.a].png' corresponds to the confusion matrix 'a' in figure 5.

Repository Organization

The organization of the rest of this repository is the following:

• IO500_prelim contains the data and execution scripts used in our preliminary quantitative analysis of IO500 under various types of interference
  • data contains the raw results used as a basis to calculate the relative slowdowns represented in Table 1
  • scripts contains the series of scripts used to generate the above mentioned data. The process is launched by the background_multi_interference.sh script which launches the other scripts during its exectution
• enzo_prelim contains the data and execution scripts used in our prelimary quantitative analysis of Enzo under various types and levels of interference
  • data contains the unprocessed (.Darshan) and processed (.csv) trace data from each exection reprsenented in Figure 1.
  • scripts contains the scripts used to generate the above mentioned traces.
    • run_enzo_interference_levels.py is the script used to run the enzo workload under various levels of cross-application interference.
    • run_enzo_interference_types.py is the script used to run the enzo workload under various different types of cross-application interference.
• NN_Model contains the training/testing data, training execution scripts, and model file for the trained model corresponding to Figures 3-5
  • data contains the raw data collected from each application split into train and test sets for model training and evaluation
  • model contains the trained model .pkl file
  • scripts contains the model training and evaluation script