Kigland OpenHPC

A lightweight workload scheduler. Similar to Slurm, but simpler and easier for small-size distributed HPC cluster.

Overview

What we have done:

• Multi GPU Assignment
• Safe Environment environment for different users
• Multi port access (HTTP, SSH, ...) (Simple Proxier)
• Web Portal

TODO:

• Job Queue
• Job Scheduler (SJF/FCFS)
• User Portal
• Multiple Physical Node Coordination
• ...

Architecture

flowchart TB
  M[/Coordinator\]
  P[/Proxier\]
  U([Users])
  subgraph HPC Cluster
    HPC1[HPC 1]
    HPC2[HPC 2]
    HPC3[HPC 3]
  end
  M --> HPC1
  M --> HPC2
  M --> HPC3
  P --> HPC1
  P --> HPC2
  P --> HPC3
  M --> P
  U -->|Submit| M
  U -->|Access| P

Loading