Tech Tip #43: Consolidate underused servers

Moving VMs off half-empty hosts can cut energy by up to 22%.

Video created by explainapaper.com

The problem

Workloads spread one-per-server or across many lightly loaded hosts keep every host powered and cooled. Over time placements drift and utilisation falls further.

The fix you can use today

  • Find hosts that are consistently underloaded.
  • Migrate their workloads onto fewer hosts, leaving headroom for peaks.
  • Power down or repurpose the emptied hosts, and repeat periodically.

One study found that packing virtual machines onto fewer servers can cut energy by up to 22%. GreenCode's research review draws on this work.

Tools that help

  • Kepler: exports energy estimates per pod and container on Kubernetes, using eBPF and hardware counters
  • Scaphandre: measures the power drawn by each process on a Linux server and exports it to Prometheus and other monitoring stacks
  • ipmitool: reads a server's power from its management controller (`ipmitool dcmi power reading`)

How GreenCode helps

GreenCode helps tackle this by mining a running estate and rebuilding it as infrastructure as code. It traces energy and CO2e to the services that cause it, including the ones barely running at all. More on this.

Tags

#right-sizing#idle#data-centre

Sources

Transcript

Moving virtual machines off half-empty hosts can cut energy by up to twenty-two percent. Many workloads sit one per server, or spread across lightly loaded hosts. Every one of those hosts stays powered, and cooled. Over time, placements drift and the hosts get emptier. Here is the fix you can use today. Find the hosts that are underloaded most of the time. Tools like Kepler, Scaphandre and ipmitool show power per container, process and server. Move their workloads onto fewer hosts. Leave room for the peaks. Then power down the empty hosts, or give them other work. And repeat this every so often. One study found this packing can cut energy by up to twenty-two percent. It also means less cooling and fewer machines to buy. GreenCode's research review draws on this work. GreenCode helps tackle this by mapping your running systems and rebuilding them as code. It shows which services use the energy, and which barely run. That was GreenCode Tech Tip number forty-three. Follow for the next tip, and find more on greencode.ai/tech-tips.

All GreenCode Tech Tips · All GreenCode videos

  • Tip #43
  • For Ops and infrastructure
  • Length 1:08
  • Category Cloud and infrastructure
GreenCode

Want this checked on your own system? Talk to the GreenCode team.

Get in touch