Petlove

As Petlove’s Google Cloud compute spend increased, GKE became the main target for optimization

Petlove’s compute usage spans both AWS and Google Cloud, the company adopting a multi-cloud strategy to ensure that their environments and workloads are always running in the most suitable cloud. However, this meant that cost reduction strategies would have to incorporate multiple complex cloud environments.

To help combat costs in their AWS account, Petlove have used Spot to run their environments on EC2 Spot Instances for quite some time, with their Google Cloud environments being the natural next progression for cost optimization. For Petlove, the main target for cost reduction in their environment was Google Kubernetes Engine (GKE).

Google Preemptible VMs

Preemptible VMs, just like their AWS counterpart spot instances, are how Google sells its spare compute capacity at a greatly reduced rate. Preemptible VMs differ from spot instances in a few key ways. One of these is that they can only run for 24 hours before the preemptible VM is terminated. This means that complex production workloads, such as Petlove’s GKE environments, are extremely difficult to run robustly on preemptible VMs. Also, preemptible VMs are not covered by SLA which makes it more challenging to use in a production workload.

Spot support for GKE provides over 70% cost reduction with no architectural changes

After using Elastigroup by Spot’s support for GKE to run Petlove’s GKE environment on preemptible VMs, Petlove swiftly experienced a significant drop in their compute costs. “It was not only cost-effective, it was also extremely simple to deploy” said Fernando Tralci, Head of Infrastructure.

Once Spot was integrated with the GKE environment, it was able to not only scale the infrastructure on preemptible VMs, but also read the pending pods and scale the appropriate preemptible instance type and size into the Kubernetes cluster, based on its Autoscaler for Kubernetes. This meant that all the scaling for their GKE environment was being automated by Spot in such a way as to ensure the best possible optimization.

Not only were Petlove using the most cost effective VM type available on Google, they were also using fewer, right-sized VMs inside their GKE environments.

“Spot’s support for GKE is great, we’re able to both optimize our VM purchasing method and simultaneously make sure that we’re only buying and using exactly what we need with as little wasted capacity as possible. That’s definitely the biggest value we could have gotten in the shortest period of time.” – Fernando Tralci, Head of Infrastructure.