Elastic Machine Pool

Maximize EKS utilization and reduce costs by 50% — automatically

Designed for DevOps/FinOps teams, EMP dramatically improves your EKS cluster CPU and memory utilization – with zero app/pod disruptions and no changes required to your app resource configuration set by the developers.

Platform9 Elastic Machine Pool dashboard shown within a laptop

Trusted by top companies

Kubernetes is the most inefficient consumer of infrastructure, especially in public clouds

Typical EKS utilization is at most 30%, which is the primary source of your compute waste.

How efficient are your EKS clusters?

A visual graph depicting a value of 30%


Real-world memory utilization metrics before EMP

Data from one of our customer’ EKS clusters: memory utilization sampled daily, averaged per month, and aggregated across all Kubernetes clusters.

Pie chart showing: 48.5% Wasted due to inefficient bin-packing; 34.4% Wasted due to high resource ‘limits’ rarely reached  in practice; 17.2% Actual utilized capacity - you want to maximize this<br />

What is the source of this inefficiency?

A diagram showing the resource requests result in Kubernetes inefficiency

Bin packing wastage

Unallocated space is created when a developer “requests” pod resources that don’t match available EC2 instance sizes. This is called  the bin-packing problem.

Allocated capacity unused by apps

To maintain app SLAs, developers often set high resource “limits”. This leads to unused resources as average usage is typically lower than peak. Kubernetes does not automatically “free”  unused resources, leading to waste.

Developers resist changes to resource config

Because they are responsible for the application’s SLA.

A diagram showing EKS resource and cost management challenges including the manually-intensive observation, monitoring, and optimization tasks

DevOps and AppDev manual loop

AppDev resist changes to pod “requests” and “limits” to  guarantee app SLA, which conflicts with the DevOps team’s need to right-size and reduce costs. This results in constant negotiations with developers and a manual loop for resource optimization.

Optimization and app disruptions

Right-sizing by DevOps often leads to pod churn (killing, moving, restarting pods), which in turn can cause app downtime and performance issues.

Elastic Machine Pool improves resource utilization, without any changes to your applications

EMP deploys an alternate virtualization layer, using AWS Bare Metal underneath, creating “Elastic VMs” (EVMs), that look and feel exactly like regular EC2 VMs.

A diagram showing the difference of EKS with & without using EMP. EMP deploys an alternate virtualization layer, using AWS bare metal underneath, creating  “Elastic VMs” (EVMs), that look and feel exactly like regular EC2 VMs

How EMP Works: resource over-provisioning

Diagram showing how EMP works to provide higher utilized bare metal capacity in aggregate
  • Behind the scenes, EVMs are optimized for best resource utilization. EMP packs more EVMs on fewer bare metal nodes, based on actual VM resource usage metrics.
  • This eliminates both the bin-packing overhead and the need to tweak your application’s resource requirements.

How EMP works: Real-time resource rebalancing

  • When the resource usage for the EVMs on a bare metal node starts going up, EMP will “live migrate” some EVMs to another Bare Metal server.
  • This is done using real-time metrics and historical usage trend pattern. Your pods stay alive without churn and app SLA is not compromised.
A diagram showing how EMP rebalances resources in real-time

Achieve over 70% utilization using EMP

EMP’s unique approach works at the virtualization and Bare Metal layers, achieving utilization gains not possible with any other existing tools or technologies.

A diagram comparing 'Before using EMP' and 'After using EMP' states shows an increase in utilization from 17.2% to 70%

Cost savings from a customer deployment

58% reduced cost
$8 million saved

A cover of the case study document titled "How a SaaS data management company slashed AWS EKS costs by 58%"

EMP is 100% compatible with EKS and AWS infrastructure

You can continue to use AWS services like EBS, EFS, and VPC. No changes are needed to existing operational tooling, automation, or upgrade processes. EMP’s EVMs can also operate in parallel to EC2 instances.

You can continue to use AWS services like EBS, EFS, and VPC. No changes are needed  to existing operational tooling, automation, or upgrade processes. EMP’s EVMs can also operate in parallel to EC2 instances.

Unlike Others, EMP optimizes at the compute layer

EMP also works side-by-side with optimization tools like Karpenter or Spot.io and cost visibility tools like CloudHealth or Cloudability, ensuring zero app disruption, and delivering an extra 35% savings.

Optimize usage via improved utilizationCheck / YesX / NoX / NoX / NoX / NoX / No
Zero pod disruptionCheck / YesX / NoX / NoX / NoX / NoX / No
Eliminate engineering and ops back and forthCheck / YesX / NoX / NoX / NoX / NoX / No
Optimize usage via bin packingCheck / YesCheck / YesX / NoX / NoX / NoX / No
Cost visibilityCheck / YesX / NoCheck / YesCheck / YesCheck / YesCheck / Yes
Resource taggingX / NoX / NoCheck / YesCheck / YesCheck / YesX / No
Unit cost allocationX / NoX / NoCheck / YesCheck / YesCheck / YesCheck / Yes

Where does EMP fit in the FinOps landscape?

  • EMP provides deeper visibility into the cost of EKS clusters, coexisting with other tools like CloudWatch.
  • You can use your existing AWS savings plans to extend to bare metal instances. EMP works alongside spot instances, ensuring SLAs for high-performance apps.
  • Unlike tools like Karpenter, there is no need to resize pods or instances, and thus there are no pod disruptions or churns. However, you can use EMP with these tools to achieve an additional 35% savings.
  • EMP is the only solution that addresses the issue of low compute utilization at the Bare Metal layer, targeting the root of the problem.
VisibilityCost and consumption
FinancialSpot instances, reserved instances, savings plan
Right-sizingPods and instances
Compute optimizationElastic Machine Pool (EMP)
Solves the root cause problem of low compute utilization

Technical deep-dive demo of EMP

Watch this video to see how easy it is to create EMP clusters and their seamless integration with EKS.

Additional Resources

Announcing Elastic Machine Pool: The most cost-effective Compute Engine for EKS

Announcing Elastic Machine Pool: The most cost-effective Compute Engine for EKS

How a SaaS data management company slashed AWS EKS costs

How a SaaS data management company slashed AWS EKS costs by 58%

The preview image for the EKS Cost Optimization Guide resource shows a collage of a monetary concept over a technology background

EKS Cost Optimization Guide – steps to optimize costs with EKS

The preview image for the Elastic Machine Pool Datasheet shows a close-up image of a coil or spring

Elastic Machine Pool datasheet — Understand how EMP works

Frequently Asked Questions

What is Platform9 Elastic Machine Pool (EMP)?
Elastic Machine Pool (EMP) is a new computing engine designed for AWS EKS that can double resource utilization while lowering costs by up to 50%. Its patent-pending software uses proven server consolidation principles and modern cloud infrastructure to dynamically scale compute instances according to actual resource usage, without impacting the application SLAs.
What results can I expect from implementing EMP?
Within weeks, most customers’ cluster utilization increases from 15-30% to more than 50%.
How do I use EMP?
EMP integrates seamlessly with your existing EKS environment. Just provide your AWS credentials, and within minutes, you can install EMP and configure it to work with your existing, or new EKS clusters. You can then direct some or all of your workloads to run on EMP, providing you with cost savings.
How does EMP help DevOps teams and Developers?
Application developers frequently define the resource requirements of their workloads based on peak CPU and memory requirements. However, real-world usage may frequently vary or be significantly lower than configured values for the majority of the time, necessitating ongoing resource parameter adjustments by DevOps teams or Developers. EMP intelligently detects actual usage and automatically fine-tunes computing resources to meet the needs of the application, eliminating the need for manual configurations.

Developers can continue to base their application’s resource requirements on peak usage. At the same time, operations teams gain a powerful lever to rein in expenses.

Does EMP require changes to my applications?
No, EMP performs its optimization automatically in the background without any app changes or impact on SLAs.
Will EMP impact my application SLAs?
No. One of the key value propositions of EMP is that it guarantees your application’s SLA 100% of the time. EMP fully understands the resource request and limit values set by your Developers for your applications. EMP combines this information with the application’s actual usage to dynamically adjust the amount of resources given to your application. So when your application actually needs the peak resources it’s configured to use, it will always be given those resources. Hence your application SLA will never be impacted with EMP.
What Kubernetes platforms does EMP support?
EMP currently only integrates with and supports AWS EKS clusters—support for other platforms is coming soon.
Why should we consider EMP when we are already using other cost management tools?
There are several monitoring and visibility tools available on the market that claim to provide significant cost savings for your Kubernetes environment. CloudHealth, Cloudability, and Datadog are great tools for understanding your cloud costs and identifying areas where you can save money. However, they do not specifically focus on automating the changes needed to increase your overall resource utilization.

Then there are other tools that make it easy to use spot instances with your EKS clusters to reduce your costs. However, spot instances are not a good fit for many applications that can not tolerate node interrupts, thus putting a ceiling on the amount of savings such tools can offer.

As a result, despite the use of these tools, Kubernetes utilization remains shockingly low, ranging between 15 and 30%. That means 70% or more of your k8s cloud resources are being completely wasted. This inefficiency is silently swelling costs to unsustainable levels.

With EMP, there is no more manual tweaking or negotiating with engineers. Let the engineers set the requests and limits on the resources that they desire, and EMP will work behind the scenes to optimize your EKS resource utilization based on demand.

EMP is complementary to your existing tools and can provide value on top of these technologies by maximizing utilization on the fly.

Can we test EMP ourselves?
We encourage you to validate the EMP value proposition through a low-risk POC. The best way to get started is to spin up a new EKS cluster or take an existing non production EKS cluster and enable the EMP compute engine for that cluster, then start deploying a small portion of your workloads using EMP. This will enable your teams to see real utilization gains without committing production clusters. No changes are required to your existing EKS clusters, tooling, or workloads.

Testing EMP via a POC is the best way to experience gains in utilization and hence, cost savings in your specific environment.

How does Platform9 work with us to enable EMP in our environment?
Once your POC is successfully complete and you validate the resulting cost savings, our team will work hand-in-hand with your teams to implement EMP in production. They will partner with you to assess your environment, evaluate savings potential, plan deployment, and ensure you extract maximum value from EMP. Consider them an extension of your team.
How does EMP handle security and access controls?
EMP adheres to strict zero-trust principles and integrates with your native AWS IAM policies and permissions.
What Kubernetes expertise is required to run EMP?
Any Kubernetes administrator can implement EMP. While implementing EMP does not require deep Kubernetes expertise, a basic understanding and knowledge of Kubernetes concepts and an in-depth understanding of your existing EKS environment are required.
Does EMP address cluster sprawl?
EMP does not directly address the cluster sprawl problem, as it does not provide tooling to reduce the number of your EKS clusters or to manage those clusters. However, by improving your overall cluster utilization, EMP allows you to significantly consolidate your overall resource footprint.
What support does Platform9 provide?
Your success is ensured by our 24×7 support, professional services, and solution engineers. Our support team consists of CKA certified experts with proven track record of excellence in managing thousands of Kubernetes clusters, and has received 100% customer satisfaction ratings four years in a row. 
I have more questions!
Reach out anytime at contact-emp@platform9.com. We are happy to answer your questions.

Ready to take action?

Meet with our founders (CTO and VP of Product) who built this product and solution experts to jointly explore the following areas:
  • Your current EKS infrastructure footprint, use cases, and workload characteristics.
  • Your organization’s Kubernetes efficiency goals.
  • Your EKS cluster CPU and memory utilization metrics.
  • How EMP can be used in your environment to achieve your desired outcomes, with a deeper dive into technical considerations such as networking, storage, performance, etc.
  • See the EMP product in action, including integration with EKS, cost-saving, and utilization dashboards.

The browser you are using is outdated. For the best experience please download or update your browser to one of the following: