Whether Public or Private Cloud, there are choices that can be made to optimise compute.
FinOps has some tried and tested recommendations for Public Cloud:
- Right-sizing
- Utilizing spot instances
- Investing in reserved instances
- Identifying and eliminating unused resources
- Consolidating idle resources
Numbers 1, 4 and 5 are also key in the Private Cloud. But looking at what’s under the hood, what’s running those VMs, can also optimise your compute.
AMD processors are available in Public Cloud and you can obviously select them for a Private Cloud. Choosing AMD CPU can result in marked benefits from reduced server count, reduced power, cooling and rack space, reduced support and licensing, and reduced networking. All of these benefits may sound like they’re only relevant to Private Cloud but when you look at them from an ESG sustainability perspective then your consumption of Public Cloud hardware is important and AMD is an option for the compute you select.
The Standard Performance Evaluation Corporation (SPEC) is a non-profit corporation formed to establish, maintain and endorse standardized benchmarks and tools to evaluate performance and energy efficiency of computing systems. SPEC benchmarks measure performance by running a set of applications with a specific configuration and workload that then enables you to compare one computer system to another – e.g. An AMD server with an Intel based one.
Taking SPEC results, its obvious that you can run more workloads on an AMD server than alternatives. So the AMD server can host more VMs or Kubernetes nodes. That means you’ll need less servers resulting in reduced power, cooling, rack space, support, licensing and networking. Obvious benefits in a Private Cloud. But its also worth selecting this in Public Cloud as it aligns to companies’ sustainability goals, especially as AMD CPUs are very power efficient.
A discussion on optimising compute wouldn’t be complete without addressing AI. With the large quantity of compute resources required to run AI, sizing of AI clusters and scoping requirements is critical from a cost, performance and sustainability perspective. Understanding the AI workload is critical to assess whether its CPU or GPU heavy. Then the choice of AI model is critical with its number of parameters, being in the billions, impacting sizing. Then the requirements for the output also influence the amount of compute – the input character size, precision factor, and response time. It’s a complicated task that if done badly will result in over-sized compute resulting in excessive spend, space, power, cooling, support and licensing. Again, if the Public Cloud is used, all these factors and results are important from both a cost and ESG sustainability perspective.
Finally, optimisations can be achieved through the location a workload is hosted in. The majority of companies run a Hybrid Cloud estate. Workloads in the Public Cloud are costing more than anticipated. Should you then refactor that workload to Public Cloud native services? Is the effort worth it? Or by simply moving the workload to a Private Cloud you can reduce cost easily but a correct assessment of value needs to be conducted first.
There’s lots that can be done to optimise compute in any Cloud with the business benefits potentially substantial. Application of FinOps principles is where to start but looking deeper beyond the well known recommendations can pay dividends.
Want to understand more how Natilik can help you on your FinOps journey? Contact us today.
Nigel Pyne
Principal Architect, Natilik
Return to Resources