Kerrighed

Five years on Kerrighed, a Single-System Image cluster operating system: OS integration, a 256-node experimental cluster, a cluster management suite, and HPC solution architecture for customers.
Completed Linux High Performance HPC Clusters Distributed Systems Operating Systems Debian SNMP Python Open Source Kerrighed

Context

Kerrighed is a Single-System Image operating system: a cluster of machines made to look and behave like a single large SMP machine, so that applications run on it unmodified. Born as research at Inria, it was industrialized by the Kerlabs spin-off – and I followed it across both.

What I did

Inria, HPC Engineer (2005 – 2007)

  • Kerrighed OS integration on Debian, CentOS and the OSCAR HPC suite of tools.
  • Release manager of the OSCAR HPC suite of tools.
  • Testbed cluster platform setup and operations.

Kerlabs, architect and developer of HPC systems (2007 – 2010)

  • Setup and operations of a 256-node experimental Kerrighed cluster.
  • Designed and implemented a cluster management suite: resource allocation, network management, A/C monitoring and control (SNMP, Python, web).
  • OS integration across Cray CX1, liquid-cooled super workstations and standard clusters, on Debian, XtreemOS, CentOS and Mandriva.
  • HPC solution architecture for customers.

Outcome

A research operating system taken all the way to customer hardware, plus the publications that came with it – Clustering the Kernel and OSCAR KernelPicker: Handling Clients Kernels.