Skip to content

Latest commit

 

History

History
52 lines (45 loc) · 1.93 KB

SETUP.md

File metadata and controls

52 lines (45 loc) · 1.93 KB
Raw

We provide the code/scripts to setup Lotus experiment in the cloudlab testbed using a c4130 node available in Wisconsin cluster.

The following instructions are for setting up/installing software dependencies on the c4130 node. The experiments are performed on the ImageNet dataset for the Image Classification task.

Setup steps

  1. Use our Lotus_c4130_cloudlab.profile as the geni script for cloudlab experiment profile. This profile uses a long term cloudlab dataset as a mounted remote filesystem with 916 GB storage. Mounted on /mydata.

  2. Grow root partition of c4130 node on cloudlab (If this step is not performed, there will be no space left in root for installations):

    export RESIZEROOT=100
sudo scripts/cloudlab/grow-rootfs.sh
export RESIZEROOT=

  3. Install Intel VTune:

    sudo bash scripts/cloudlab/install_vtune.sh
# Make sure to run below commands 
echo "source /opt/intel/oneapi/setvars.sh" >> ~/.bashrc
sudo usermod -aG vtune $USER && newgrp vtune

    Note: we used Intel(R) VTune(TM) Profiler 2023.2.0 (build 626047)

  4. Install CUDA 11.8:

    sudo bash scripts/cloudlab/cuda_installer_script.sh

  5. Install CuDNN 8.7.0:

    sudo bash scripts/cloudlab/cudnn_installer_script.sh

  6. Reboot the machine:

    sudo reboot

  7. Install vmtouch:

    sudo apt install vmtouch

  8. Install anaconda from here:

    curl -O https://repo.anaconda.com/archive/Anaconda3-2024.06-1-Linux-x86_64.sh
# Follow instructions (add Anaconda to `/mydata/iiswc24/anaconda3` when prompted)
bash Anaconda3-2024.06-1-Linux-x86_64.sh
source ~/.bashrc

  9. Download and setup ImageNet dataset:

    bash scripts/cloudlab/download_imagenet.sh