EKS Cost Optimization From Platform9

Platform9 returned to Cloud Field Day in 2024 several years after their last appearance. Their product has undergone significant changes since they last presented so it almost feels like a relaunch of sorts.

Company Overview

Platform9 was founded in late 2013 with the mission to democratize cloud computing. Their flagship product is focused on Private and Edge clouds. It is built on popular open source projects leveraging Platform9 IP. Once the platform is racked, stacked, and powered on, Platform9 takes over management of the infrastructure and customers just consume it.

From an operational perspective, Platform9 provides a 99.9% SLA with what they call "Proactive Ops". They will as a company collect data 24/7 to ensure consistent operations and respond to infrastructure incidents. Customers can allow Platform9 to collect logs proactively if they wish to share additional date with Platform9.

EMP (Elastic Machine Pool) Intro and Overview

The focus of Platform9's presentation at Cloud Field Day however, was their new Elastic Machine Pool (EMP) solution. EMP is a new compute engine meant to optimize cloud Kubernetes costs in the cloud. Although there are multiple tools on the market focused on solving the problem of K8S spend on the market, Platform9 believes they have a unique solution. According to Platform9, typical EKS utilization is only 30% of the available compute capacity, which creates a lot of wasted capacity. Most of this inefficiency is due unoptimized environments by customers without the proper expertise to do so.

Much like the classic overprovisioned VMs in VMware clusters, many applications are deployed with the max resources they may need instead of something closer to their average utilization. This is an age old dilemma that infrastructure teams face when developers and application teams request resources for their worst case scenario, leaving the majority of compute capacity wasted and unused.

This is essentially history repeating itself. We have seen it over and over with app teams requesting huge VMs or DBAs making sure they get the most RAM possible. EKS is just a new place for this problem to manifest itself. Existing finops solutions attempt to solve the problem of cost optimization without truly addressing compute optimization, which is a large cost component. This is the value prop of EMP.

In a nutshell, EMP is designed to allow customers to overprovision K8S applications on top of AWS bare metal, similarly to how VMware allowed customers to overprovision compute resource in vSphere Clusters.

As resources utilization rises, EMP can scale and migrate nodes without disrupting the applications that are running on them.

Overall I'm not super knowledgeable about Kubernetes and was out of my depth pretty quickly during the discussion. However, it became clear through some comments from my fellow delegates that this is a tool meant to solve an organizational problem.

I agree that we would all like to see everyone take more responsibility for their applications and that things would be much simpler if people understood how the infrastructure they are using works. But I have been fighting that battle for my whole career. In an ideal world we wouldn't have to fight the battle over application optimization in a never ending loop. But until we do (if we ever do) tool focused on cost reduction will have a place in the market and Platform9 appears to have carved out a niche.