SniperSim Simulator for Toleo

This is a sniper sim and DRAMSim based simulator for Toleo (ASPLOS '24) Please refer to our paper "Toleo: Scaling Freshness to Tera-scale Memory Using CXL and PIM" for details.

Getting Started

Running evaluation for Toleo requires:

1. Install DRAMSim3 from this fork.
2. Install simulator sniper-toleo. (this repo)
3. setup benchmarks with PIN hooks. Forks with PIN hooks can be found in this list
4. Download simulation script run_toleo_sim.py for automated simulation on given benchmarks.

File Structure

toleo_root
├── run_toleo_sim.py
├── DRAMsim3
├── sniper-toleo
│   ├── DRAMsim3 -> ../DRAMsim3 (link)
│   ├── run-sniper
│   └── README.md (this file)
├── genomicsbench
│   ├── input-datasets
│   │    ├── bsw
│   │    ├── chain
│   │    ...
│   └── benchmarks
│      ├── fmi
│      │  └── sim-<date-time>/<region>/sim.out
│      ├── bsw-s
│      ...
├── gabps
│   └── run
│      ├── pr-kron-s
│      │  └── sim-<date-time>/<region>/sim.out
│      ├── sssp
│      ....
├── redis
├── memtier_benchmark
├── memcached
├── hyrise
└── llama2.c

Make TOLEO_ROOT directory

mkdir toleo_root
cd toleo_root

Install DRAMsim3

git clone https://github.com/joydddd/DRAMsim3
cd DRAMsim3

Please follow the README from this fork to build DRAMsim3. THERMAL is not required for toleo simulation. Run DRAMsim3 test to make sure it is successfully installed:

 ./build/dramsim3main configs/DDR4_8Gb_x8_3200.ini -c 100000 -t tests/example.trace
cat dramsim3.json

Install sniper-toleo

Prerequisite

• A DRAMsim3 installation from this fork, and link to sniper-toleo repo.

## in folder toleo_root. 
git clone [email protected]:joydddd/sniper-toleo.git # clone sniper-toleo (this repo)cd sniper-toleo
ln -s ../DRAMsim3 DRAMsim3

(optional) Setup docker environment

cd sniper-toleo/docker
make   # build docker image
make run # run docker as user
# make run-root # runs docker as root.
cd .. 

Now you're in the docker container. Continue to build sniper-toloe following the steps below. You could skip the dependent package installation.

Install SniperSim

Please follow the naive install instructions on Sniper Sim Getting Started Page to install sniper-toleo. Necessary steps are provided below

1. Install dependent packages.

sudo dpkg --add-architecture i386
sudo apt-get install binutils build-essential curl git libboost-dev libbz2-dev libc6:i386 libncurses5:i386 libsqlite3-dev libstdc++6:i386 python wget zlib1g-dev

Note

Know issue: snipersim assum python2 as the default python version. Make sure your python command points to python2.7.

2. build simulator with PIN

make USE_PIN=1 -j N #where N is the number of cores in your machine to use parallel make

Test Run

cd test/fft
make

This command runs sniper-toleo simulation for three setups: Toleo, C+I, and no-protections. Simulation results can be found in folders test/fft/no_protection_output test/fft/ci_output test/fft/toleo_output respectively.

Run Your Own Benchmark

Now you've successfully setup sniper-toleo and is ready to simluate toleo for any benchmark of your choice! run the simulation with

./run-sniper -n 32 -c <arch> -d <output-dir>  -- <run benchmakr cmd>
# <arch> = zen4_cxl: baseline no protection, 
#          zen4_vn:  toleo
#          zen4_no_freshness: CI
#          zen4_cxl_invisimem: InvisiMem (Please use invisimem branch of sniper-toleo for this architecture!) 
#          zen4_no_dramsim: light weighted Toleo without DRAMSim3. Used for Toleo page type analysis. 

More controls over the simulation are available via ./run-sniper --help.

Note

Our simulator is configured for 32 cores. Please setup your benchmark to run with 32 threads for best experience. Benchmarks with >32 threads might run into deadlock problems.

Citing Toleo

If you use Toleo in your research, please cite:

@misc{dong2024toleoscalingfreshnessterascale, title={Toleo: Scaling Freshness to Tera-scale Memory using CXL and PIM}, author={Juechu Dong and Jonah Rosenblum and Satish Narayanasamy}, year={2024}, eprint={2410.12749}, archivePrefix={arXiv}, primaryClass={cs.AR}, url={https://arxiv.org/abs/2410.12749}, }

Additional Information

Benchmarks Used in Evalutions

We provided PIN hooks inserted benchmarks used in our evaluations in this list. Please refer to benchmark setup instruction.

Evaluation Workflow

We provided a batch run script to run our evaluation workflows. Please refer to Evalution Workflow Instructions.