About this repository

This repository contains the gem5 simulator code for evaluating the design idea proposed in ASAP: A Speculative Approach to Persistence. ASAP is implemented with gem5 v20.0.0.3. There are 4 branches in this repository, each for a different model (baseline, ASAP, HOPS, ideal).

Usage guide

This readme aims to help in the installation, compilation and execution of this simulator. The repository contains scripts to reproduce the results presented in the paper.

Installing

1. Install gem5 dependencies.

   sudo apt install build-essentialgit m4 scons zlib1g zlib1g-dev libprotobuf-dev python python-dev protobuf-compiler libgoogle-perftools-dev libprotoc-dev libboost-all-dev pkg-config

2. Clone this repository.

   git clone https://github.com/multifacet/ASAP.git

   We use full system simulations to evaluate ASAP. gem5 full system simulations require a kernel image and disk images with the workloads. The kernel and disk images used for evaluating ASAP are available at: http://pages.cs.wisc.edu/~sujayyadalam/asap/

3. Download the kernel image into gem5 directory and the disk images to a directory called disks.

   cd ASAP
   wget http://pages.cs.wisc.edu/~sujayyadalam/asap/vmlinux_12
   mkdir disks; cd disks;
   wget -r -np -nd -A "*.img" http://pages.cs.wisc.edu/~sujayyadalam/asap/images
   cd ..

4. While running on Ubuntu 20.04, certain Python 2 modules needs to be installed. For this, we would require pip2.

   wget https://bootstrap.pypa.io/pip/2.7/get-pip.py
   python2 get-pip.py

   Install six module.

   python2 -m pip install six

Building gem5

1. gem5 requires GCC version 7.0 or greater and scons >= 3.0. gem5 v20.0.0.3 (version of this repository) has some issues with Python 3. So use Python 2 to compile gem5.

   python2 $(which scons) build/X86/gem5.fast -j<threads>

   For more information about building gem5 refer to http://www.gem5.org/documentation/general_docs/building

2. If you wish to change the model, then remove the build folder before re-compiling.

   git checkout <branch>
   rm -r build
   python2 $(which scons) build/X86/gem5.fast -j<threads>

Evaluating workloads

The disk images mentioned above include all the workloads used for evaluating ASAP. If you wish to run these workloads, you could use the run.sh script.

./run.sh <workload>

If you are evaluating ASAP or HOPS, an additional parameter for the persistency model [arp|epoch] needs to be passed.

./run.sh <workload> <persistency_model>

If you wish to run all the workloads in parallel, you could use the run_all.sh script. Note that this would require atleast 16 cores and 150GB for memory.

./run_all.sh

Reproducing results from the paper

To reproduce the performance comparison results presented in the graph, you could use the reproduce_results.py. Invoke this script after you have completed all the experiments, i.e. all workloads with all models.

./reproduce_results.py

License

This artifact uses the same license as gem5. See LICENSE for more information.