python3, numpy, scipy, sortedcollections
Please refer to our paper accepted at EUROSYS 2020
start_simulations.py is the start point from which you can run various rack scale simulations. It accepts multiple arguments, but you can generate the large scale simulation results we presented in our paper (Figure 7, 8 and 9) with the default configuration:
python3 start_simulations.py
This would run the default large-scale simulation where the amount of far memory and additional local memory vary, the results would be written in a text file stored in results/results_192G_48cores.
| Argument | Description |
|---|---|
| --num_random, -n | Number of randomly generated workloads. |
| --limits, -l | Limits of m2c. |
| --cpu, -c | Number of cpu per machine. |
| --mem, -m | Amount of memory per machine (unit is MB). |
| --jps, -j | Number of jobs per server. |
| --filename, -f | Filename for final results. |
| --simu_name, -s | Name of the simulation loop function. |
| --use_small_workload | To use small workload. |
simulation_one_time.py allows you to run single simulation with small workloads. test.sh contains examples usage of simulation_one_time.py. test.sh requires two parameters: seed to generate workload and amount of far memory. Here is an example usage:
./test.sh 2000 32768
For additional questions please contact us at cfm@lists.eecs.berkeley.edu