QPRAC (HPCA 2025)

Introduction

This is the code artifact for the paper "QPRAC: Secure and Practical PRAC-based Rowhammer Mitigation using Priority Queues", presented at HPCA 2025 and received the Distinguished Artifact Award.

Authors: Jeonghyun Woo (University of British Columbia), Shaopeng (Chris) Lin (University of Toronto), Prashant J. Nair (University of British Columbia), Aamer Jaleel (NVIDIA), Gururaj Saileshwar (University of Toronto).

You can reproduce our security and performance evaluations as follows.

Acknowledgement

The performance evaluation infrastructure in this artifact has been adapted from SRS (HPCA 2023) and BreakHammer (MICRO 2024).

Requirements

Run-time Environment: We suggest using a Linux distribution compatible with g++-10 or newer for the performance evaluations. For example, Ubuntu 22.04 or later is recommended if you prefer Ubuntu. This artifact has been tested on Ubuntu 22.04 and Rocky Linux 9.4.

Security Evaluations:

Python3 (Tested on V3.11.5)

Performance Evaluations:

Software Dependencies:
- g++ with C++20 support (tested with version 12.4.0).
- Python3 (recommended: version 3.10 or above).
Hardware Dependencies:
- 40 core system with at least 128GB memory is recommended to run all benchmarks efficiently.

Security Analysis

Please run the following steps to regenerate the security analysis and figures:

Clone the artifact and run the code.

Fetch the code: git clone https://github.com/sith-lab/qprac.git
Run Artifact with regeneration of data (monte-carlo analysis) (slower - 2 hours):
```
  $ cd qprac/security_analysis
  $ bash ./run_artifact.sh
```

Run Artifact without regenerating data

  $ cd qprac/security_analysis
  $ bash ./run_artifact.sh --use-sample

These scripts execute the following commands outlined below.

Steps Run By Script

Below are the steps run by our above script in an automated manner.

Monte Carlo Analysis (for Max R1): This obtains the Max R1 value without and with Proactive Mitigation, used for Fig-7 and Fig-11 respectively.

  $ cd qprac/security_analysis/analysis_scripts
  $ python3 equation3.py <MIN_WAVE_LEN> <MAX_WAVE_LEN> <RESULT_EQ3>
  $ python3 equation3_pro.py <MIN_WAVE_LEN> <MAX_WAVE_LEN> <RESULT_EQ3_PRO>

Figure 2. Attack t-bit toggling:

  $ cd qprac/security_analysis/figure2
  $ # Example: 
  $ #   python3 ../analysis_scripts/tbit_attack.py 6 8 10 > tbit_attack.txt
  $ #   python3 figure2_plot.py 3 tbit_attack.txt
  $ python3 ../analysis_scripts/tbit_attack.py <T_Bit> ... > tbit_attack.txt
  $ python3 figure2.py <TESTED_BITS> tbit_attack.txt

Figure 6. Maximum N_Online Equation (2):

  $ cd qprac/security_analysis/figure6
  $ # Example: 
  $ #   python3 ../analysis_scripts/equation2.py 0 $((2**17)) > PRAC1-4.txt
  $ #   python3 figure6_plot.py PRAC1-4.txt $((2**17))
  $ python3 ../analysis_scripts/equation2.py <MIN_WAVE_LEN> <MAX_WAVE_LEN> > <RESULT_EQ2>
  $ python3 figure6.py <RESULT_EQ2> <MAX_WAVE_LEN>

Figure 7. Maximum R_1 from equation (3):

  $ cd qprac/security_analysis/figure7
  $ # Example: 
  $ #   python3 ../analysis_scripts/equation3.py 0 $((2**17)) R1.txt
  $ #   python3 figure7_plot.py R1.txt
  $ python3 ../analysis_scripts/equation3.py <MIN_WAVE_LEN> <MAX_WAVE_LEN> <RESULT_EQ3>
  $ python3 figure7_plot.py <RESULT_EQ3>

Figure 8. Maximum TRH:

  $ cd qprac/security_analysis/figure8
  $ # Example:
  $ #   python3 ../analysis_scripts/equation2.py 0 $((2**17)) > PRAC1-4.txt
  $ #   python3 ../analysis_scripts/equation3.py 0 $((2**17)) R1.txt
  $ #   python3 figure8_plot.py PRAC1-4.txt R1.txt
  $ python3 ../analysis_scripts/equation2.py <MIN_WAVE_LEN> <MAX_WAVE_LEN> > <RESULT_EQ2>
  $ python3 ../analysis_scripts/equation3.py <MIN_WAVE_LEN> <MAX_WAVE_LEN> <RESULT_EQ3>
  $ python3 figure8.py <RESULT_EQ2> <RESULT_EQ3>

Figure 11. Maximum R_1 with or without Proactive Mitigation:

  $ cd qprac/security_analysis/figure11
  $ # Example:
  $ #   python3 ../analysis_scripts/equation3.py 0 $((2**17)) R1.txt
  $ #   python3 ../analysis_scripts/equation3_pro.py 0 $((2**17)) R1_PRO.txt
  $ #   python3 figure11_plot.py R1.txt R1_PRO.txt
  $ python3 ../analysis_scripts/equation3.py <MIN_WAVE_LEN> <MAX_WAVE_LEN> <RESULT_EQ3>
  $ python3 ../analysis_scripts/equation3_pro.py <MIN_WAVE_LEN> <MAX_WAVE_LEN> <RESULT_EQ3_PRO>
  $ python3 figure11_plot.py <RESULT_EQ3> <RESULT_EQ3_PRO>

Figure 12. Maximum N_Online with vs without Proactive Mitigation:

  $ cd qprac/security_analysis/figure12
  $ # Example:
  $ #   python3 ../analysis_scripts/equation2.py 0 $((2**17)) > PRAC1-4.txt
  $ #   python3 ../analysis_scripts/equation2_pro.py 0 $((2**17)) > PRAC1-4_PRO.txt
  $ #   python3 figure12_plot.py PRAC1-4.txt PRAC1-4_PRO.txt $((2**17))
  $ python3 ../analysis_scripts/equation2.py <MIN_WAVE_LEN> <MAX_WAVE_LEN> > <RESULT_EQ2>
  $ python3 ../analysis_scripts/equation2_pro.py <MIN_WAVE_LEN> <MAX_WAVE_LEN> > <RESULT_EQ2_PRO>
  $ python3 figure12.py <RESULT_EQ2> <RESULT_EQ2_PRO> <MAX_WAVE_LEN>

Figure 13. Maximum TRH with or without Proactive Mitigation:

  $ cd qprac/security_analysis/figure13
  $ # Example:
  $ #   python3 ../analysis_scripts/equation2.py 0 $((2**17)) > PRAC1-4.txt
  $ #   python3 ../analysis_scripts/equation2_pro.py 0 $((2**17)) > PRAC1-4_PRO.txt
  $ #   python3 ../analysis_scripts/equation3.py 0 $((2**17)) R1.txt
  $ #   python3 ../analysis_scripts/equation3_pro.py 0 $((2**17)) R1_PRO.txt
  $ #   python3 figure13_plot.py PRAC1-4.txt PRAC1-4_PRO.txt R1.txt R1_PRO.txt
  $ python3 ../analysis_scripts/equation2.py <MIN_WAVE_LEN> <MAX_WAVE_LEN> > <RESULT_EQ2>
  $ python3 ../analysis_scripts/equation2_pro.py <MIN_WAVE_LEN> <MAX_WAVE_LEN> > <RESULT_EQ2_PRO>
  $ python3 ../analysis_scripts/equation3.py <MIN_WAVE_LEN> <MAX_WAVE_LEN> <RESULT_EQ3>
  $ python3 ../analysis_scripts/equation3_pro.py <MIN_WAVE_LEN> <MAX_WAVE_LEN> <RESULT_EQ3_PRO>
  $ python3 figure13.py <RESULT_EQ2> <RESULT_EQ2_PRO> <RESULT_EQ3> <RESULT_EQ3_PRO>

Performance Evaluations Using Ramulator2

Requirements

Software Dependencies

g++ with C++20 support (recommended: g++ version 10 or above and tested with version 11.4.0 and 12.4.0).
Python3 (recommended: version 3.10 or above).

Hardware Dependencies

Personal Server: A system with at least 128GB of memory and 40+ cores is recommended to run all benchmarks efficiently.

Steps for Performance Evaluations

1. Clone the Repository

Ensure you have already cloned the repository during the security analysis:

git clone https://github.com/sith-lab/qprac.git

2. Set Simulation Configuration Parameters

Using Slurm

Configure the following parameters in ./perf_analysis/run_artifact.sh or relevant Slurm scripts (run_slurm_fig*.sh):

SLURM_PART_NAME: Partition name for Slurm jobs.
SLURM_PART_DEF_MEM: Default memory size for jobs (recommended: ≥4GB).
SLURM_PART_BIG_MEM: Memory size for jobs requiring large memory (recommended: ≥12GB).
MAX_SLURM_JOBS: Maximum number of Slurm jobs submitted.

Using a Personal Server

Configure the following parameter in ./perf_analysis/run_artifact.sh or run_ps_fig*.sh:

PERSONAL_RUN_THREADS: Number of parallel threads to use for simulations.

3. Run the Artifact

Run the following commands to install dependencies, build Ramulator2, and execute simulations.

Note: We strongly recommend using Slurm with a cluster capable of running bulk experiments to accelerate evaluations.

Main Experiments Only (Figures 14 and 15)

Using Slurm: Faster (~16 hours on a cluster with 500+ cores).

cd perf_analysis/
bash ./run_artifact.sh --method slurm --artifact main

Using a Personal Server: Slower (~1 day on an Intel Xeon with 128GB memory).
```
cd perf_analysis/
bash ./run_artifact.sh --method personal --artifact main
```

All Experiments (Figures 14–20)

Using Slurm: Faster (~1 day on a cluster with 500+ cores).

cd perf_analysis/
bash ./run_artifact.sh --method slurm --artifact all

Using a Personal Server: Slower (~6 days on an Intel Xeon with 128GB memory).
Running all experiments on a personal server may take significant time (days to a week). Thus, we highly recommend reviewing the results for main results (Figure 14 and 15) first before proceeding with all experiments if using a personal server with limited resources (e.g., < 256GB DRAM).
```
cd perf_analysis/
bash ./run_artifact.sh --method personal --artifact all
```

4. Generate Figures

After completing simulations, use the commands below to generate plots. Alternatively, use the Jupyter Notebook (perf_analysis/plot_scripts/plot.ipynb). Generated PDFs can be found in perf_analysis/results/plots/.

Main Figures (Figures 14 and 15)

cd perf_analysis/
bash ./plot_main_figures.sh

Figures 19 to 20 (Figures 19–20)

cd perf_analysis/
bash ./plot_fig19_20.sh

All Figures (Figures 14–20)

cd perf_analysis/
bash ./plot_all_figures.sh

Detailed Steps

Prerequisites

Install Python dependencies, download required traces, and build Ramulator2:

cd perf_analysis/
bash run_prerequisite.sh

Execution

Set simulation configuration parameters:

Slurm: Configure SLURM_PART_NAME, SLURM_PART_DEF_MEM, SLURM_PART_BIG_MEM, and MAX_SLURM_JOBS in run_slurm_fig*.sh.
Personal Server: Configure PERSONAL_RUN_THREADS in run_ps_fig*.sh.

Run experiments:

Using Slurm

Main Results (Figures 14 and 15):

cd perf_analysis/
bash run_slurm_fig14_15.sh

Figure 16: Sensitivity to Number of RFMs per Alert:
```
cd perf_analysis/
bash run_slurm_fig16.sh
```
Figure 17: Sensitivity to Service Queue Size:
```
cd perf_analysis/
bash run_slurm_fig17.sh
```
Figure 18: Sensitivity to Back-Off Threshold:
```
cd perf_analysis/
bash run_slurm_fig18.sh
```
Figure 20: Comparison with In-DRAM Mitigations for Ultra-Low TRH:
```
cd perf_analysis/
bash run_slurm_fig20.sh
```

Using a Personal Server

Main Results (Figures 14 and 15):

cd perf_analysis/
bash run_ps_fig14_15.sh

Figure 16: Sensitivity to Number of RFMs per Alert:
```
cd perf_analysis/
bash run_ps_fig16.sh
```
Figure 17: Sensitivity to Service Queue Size:
```
cd perf_analysis/
bash run_ps_fig17.sh
```
Figure 18: Sensitivity to Back-Off Threshold:
```
cd perf_analysis/
bash run_ps_fig18.sh
```
Figure 20: Comparison with In-DRAM Mitigations for Ultra-Low TRH:
```
cd perf_analysis/
bash run_ps_fig20.sh
```

Collate Results

Once simulations complete, generate CSV files using the commands below. Generated csv files can be found in perf_analysis/results/csvs/.

Main Results (Figures 14 and 15):

cd perf_analysis/plot_scripts
python3 generate_csv_fig14_15.py

Figure 16:
```
python3 generate_csv_fig16.py
```
Figure 17:
```
python3 generate_csv_fig17.py
```
Figure 18:
```
python3 generate_csv_fig18.py
```
Figure 19:
```
bash generate_csv_fig19.sh
```
Figure 20:
```
python3 generate_csv_fig20.py
```

Generate Plots

After collating results, generate the plots in using the commands below. Alternatively, use the Jupyter Notebook (perf_analysis/plot_scripts/plot.ipynb). Generated PDFs can be found in perf_analysis/results/plots/.

Main Results (Figures 14 and 15):

cd perf_analysis/plot_scripts
python3 plot_fig14_15.py

Figure 16:
```
python3 plot_fig16.py
```
Figure 17:
```
python3 plot_fig17.py
```
Figure 18:
```
python3 plot_fig18.py
```
Figure 19:
```
python3 plot_fig19.py
```
Figure 20:
```
python3 plot_fig20.py
```

Name		Name	Last commit message	Last commit date
Latest commit History 70 Commits
perf_analysis		perf_analysis
security_analysis		security_analysis
.gitignore		.gitignore
README.md		README.md

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

QPRAC (HPCA 2025)

Introduction

Acknowledgement

Requirements

Security Analysis

Clone the artifact and run the code.

Steps Run By Script

Performance Evaluations Using Ramulator2

Requirements

Software Dependencies

Hardware Dependencies

Steps for Performance Evaluations

1. Clone the Repository

2. Set Simulation Configuration Parameters

Using Slurm

Using a Personal Server

3. Run the Artifact

Main Experiments Only (Figures 14 and 15)

All Experiments (Figures 14–20)

4. Generate Figures

Main Figures (Figures 14 and 15)

Figures 19 to 20 (Figures 19–20)

All Figures (Figures 14–20)

Detailed Steps

Prerequisites

Execution

Run experiments:

Using Slurm

Using a Personal Server

Collate Results

Generate Plots

About

Releases

Packages

Contributors 3

Languages

sith-lab/qprac

Folders and files

Latest commit

History

Repository files navigation

QPRAC (HPCA 2025)

Introduction

Acknowledgement

Requirements

Security Analysis

Clone the artifact and run the code.

Steps Run By Script

Performance Evaluations Using Ramulator2

Requirements

Software Dependencies

Hardware Dependencies

Steps for Performance Evaluations

1. Clone the Repository

2. Set Simulation Configuration Parameters

Using Slurm

Using a Personal Server

3. Run the Artifact

Main Experiments Only (Figures 14 and 15)

All Experiments (Figures 14–20)

4. Generate Figures

Main Figures (Figures 14 and 15)

Figures 19 to 20 (Figures 19–20)

All Figures (Figures 14–20)

Detailed Steps

Prerequisites

Execution

Run experiments:

Using Slurm

Using a Personal Server

Collate Results

Generate Plots

About

Resources

Stars

Watchers

Forks

Releases

Packages 0

Contributors 3

Languages

Packages