Skip to content

TODO List

This page lists some ideas to improve the cluster and also contains some of the current problems.

  • Make cpufp work on M1 Ultra (for now there is a compilation error)
  • On Orin boards, users should be automatically added to the video group to use OpenCL 
    sudo usermod -a -G video [USER]
  • Document each board with memory bandwidth and synthetic benchmarks
    • Optical flow: Horn & Schunck
    • Labeling: FLSL & FLSL 3D
    • CPU/RAM memory bandwidth: bandwidth
    • GPU/RAM memory bandwidth: clpeak
    • Labeling and motion detection: FMDT
    • \(n\)-body: MUrB
    • Simulation of digital communications: AFF3CT
  • Document RAID 1 configuration
    • Split the documentation in two parts: one for the users and the other for the admins
  • Document GPIOs and how to combine them with the energy measurement boards
  • Study and maybe install ansible to manage the jobs, this is a cluster management and job scheduling system like Slurm but more specific to heterogeneous setups
  • Install the module environment on the frontend (to load specific cross-compilers for instance)
  • Install cross-compilers on the frontend to target Aarch64 boards
    • GNU Compiler (C and C++) 13.2.0 (/softwares/gcc-13.2.0/)
    • Clang Compiler
  • Document and align the different installed softwares and their versions (python, gcc...)
  • Create scripts to manage the users on the frontend and the nodes
  • Install Slurm daemon (slurmd) and Munge (munge) on the nodes
    • Odroid-XU4
    • Jetson TX2
    • Jetson AGX Xavier
    • OptiPlex 7070 (Brubeck)
    • Jetson Xavier Nano
    • Raspberry Pi 4 Model B
    • Jetson Xavier NX
    • Jetson Orin NX
    • Jetson AGX Orin
    • Jetson Orin Nano
    • Orange Pi 5 Plus
    • Mac Studio M1 Ultra
  • Install NFS
  • Install Slurm control (slurmctld) on the frontend