BlueHive Tutorial

VPN Access

If you want to connect to BlueHive from outside UofR. You will need to connect to VPN first. Please check out the instruction here. If you are using UR wifi, just skip to second part.

For Linux users, there is a more convenient way to connect to UofR's VPN.

  1. Enroll in Duo two-factor authentication
  2. Install openconnect.

    sudo apt-get openconnect (Linux)
    brew install openconnect (MacOS)
  3. Connect to VPN. Note that you will be required to provide two passwords. The second one should be your Duo passcode.

    sudo openconnect vpn.rochester.edu

Using BlueHive 2.5

Once you are in UofR's network, you can then connect to BlueHive 2.5 through SSH.

ssh [your_netid]@bh25fen.circ.rochester.edu

The BlueHive system uses slurm to manage the distributed resources. Each time you want to run a program, you should NOT run it through the head node (the node you logged in through SSH) directly. Instead, you need to run your job on computation nodes. To submit a job, you may choose to use the srun or sbatch command. Here and here are some tutorials about using these commands.

One of the most convenient way of using the computation nodes is through the interactive command. This command allows you to log in to a computation node where you can then run whatever programs you like. You can use the interactive command by simply typing (from a head node of course)

interactive -p gpu-debug -t 1:00:00

You will log into a gpu model with two Tesla K20.

 

interactive -p gpu --gres=gpu:1

Note that the last option denotes the number of GPUs you want to use on the GPU node. You should always try to use just one GPU. Unless you know how to use multiple GPUs for deep learning training and your network requires a large amount of GPU memory.


 

Deep learning tools installation

  1. Install anaconda

    wget https://repo.continuum.io/archive/Anaconda2-4.4.0-Linux-x86_64.sh
    bash Anaconda2-4.4.0-Linux-x86_64.sh
  2. Install Pytorch

    conda install pytorch torchvision cuda80 -c soumith
     
  3. Install Tensorflow gpu mode 
    conda install -c anaconda tensorflow-gpu