Reducing the Energy Part of the HPC Cluster Cost Equation


By Lionel Gibbons | January 19, 2017 |



hpc clusterAccording to an award-winning paper from researchers at the Barcelona Supercomputing Center, there is a way to immediately reduce the energy costs of operating a HPC cluster. Here are the details, as reported by Inside HPC.

  • A significant contributor to the costs of operating a HPC cluster is energy, and it also is a major factor in the total cost of owning one, the report stated.
  • BSC researchers stated that up to a 52 percent savings to energy costs could be had by scaling in applications on large-memory nodes.
  • By achieving a 52 percent savings, the researchers stated, the investment made into upgrading the hardware could be recouped in less than five years.
  • Scaling in is explained as using a reduced number of applications compute nodes. The researchers proposed horizontal scaling on large-memory nodes.
  • Scaling in is the opposite of the traditional "scaling" that is known by the HPC community, which is actually "scaling out" -- which uses a larger number of application processes and compute nodes.
  • The researchers noted that the per-node memory capacity must be increased in order to maximize scale-in for energy savings.
  • The paper won Best Paper of the MEMSYS 2016 conference in November.
  • According to MEMSYS 2016 program chair, Bruce Jacob, the committee appreciated the paper because -- unlike other research -- it presented a solution that can be applied immediately and requires no changes to the system architecture.

Cost savings are a very important part of effective HPC cluster management. Bright Cluster Manager enables both cost and energy savings by automatically powering down unused nodes and system components. So, if your motivation to reduce energy consumption comes from a commitment to the planet or a commitment to the bottom line, Bright can help you achieve your goals.