Skip to content

High Performance Computing clusters at INFN-CNAF

An HPC cluster dedicated to CERN studies is available at INFN-CNAF (Bologna, Italy) allowing for MPI applications across multiple nodes. The cluster presently features a total of about 800 CPU-cores.

Access

Access to the cluster can be requested by sending an email to abp.computing[AT]cern.ch including the following information:

  • The filled in form. Please add Daniele Cesini as contact person and "Access to the hpc_acc Cluster to run High Energy Physics simulation codes" as the reason.

Usage

  • Users should connect via ssh to bastion.cnaf.infn.it and from there access the cluster front-end called ui-hpc2.cr.cnaf.infn.it.
  • The LSF system is installed on the cluster to manage job submissions. Jobs should be submitted to the ”hpc-acc” queue.
  • If you have inquiries about running your jobs on this cluster or need technical support, please write an email to hpc-support@listsNOSPAMPLEASE.cnaf.infn.it
  • A node called hpc-201-11-35 is equipped with four V100 GPUs, and can be used interactively.

To install sixtrack on hpc-201-11-35 you can source

scl enable devtoolset-7 bash
git clone https://github.com/SixTrack/SixTrack.git
cd SixTrack
sed -i 's/cmake /cmake3 /g' cmake_six
./cmake_six

A possible trick to connect to the CNAF server is to configure your ~.ssh/config (on your local machine, I tested it on my OSX and on UBUNTU) with

Host bastion
 HostName bastion.cnaf.infn.it
 User your_user_name
 ForwardX11 yes
 ForwardX11Trusted yes

Host b001
 User your_user_name
 HostName hpc-201-11-01-a
 ProxyJump bastion
 ForwardX11 yes
 ForwardX11Trusted yes
 LocalForward your_port_number_1 localhost:your_port_number_1

Host bgpu
 User your_user_name
 HostName hpc-201-11-35.cr.cnaf.infn.it
 ProxyJump bastion
 ForwardX11 yes
 ForwardX11Trusted yes
 LocalForward your_port_number_2 localhost:your_port_number_2

where you have to replace your_user_name, your_port_number_1 and your_port_number_2. The your_port_number_1 and your_port_number_2 can be guessed (e.g., typically integer >8000, if someone is already using the port will be available for you).

Then you can launch (if installed in your miniconda) jupyter lab on b001 or bgpu (first you need to connect via ssh b001 or ssh bgpu) 

jupyter lab --no-browser --port your_port_number

where your_port_number is your_port_number_1 or your_port_number_2 depending if you are connected to b001 or bgpu, respectively. If you open your jupyter server in a tmux, you can start working on your desktop at the office and continue when you TW from home.

Relevant links to obtain further information on this cluster are:

Support