Demandbase abstracts their Kubernetes infrastructure and reduces costs by 70%

The Challenge

When the Demandbase engineering team shifted to a microservices architecture for their ad bidding systems, Kubernetes was the container orchestrator of choice.

With a dynamic infrastructure deployed across 4 AWS regions, their bidding apps require bursting up to ~3000 nodes at different times all around the world.

To ensure that their proprietary, Demand-Side Platform (DSP), all its components (such as their real-time bidder), as well as their GDPR products, were always available, Micheal Waltz, Principal DevOps Engineer and Rafael Moreira, an Infrastructure Architect at Demandbase, sought out a Kubernetes-native solution that would smoothly handle all the infra management, without time-consuming, human intervention.

Rafael explained “I didn’t want the team dealing with a bunch of infra components, constantly checking that our Pods are able to run, and regularly trimming any over-provisioning.”

On the other hand, while Micheal considered Serverless options, he ultimately ruled them out as he wanted granular control of the underlying infrastructure for fine-tuned performance of Demandbase’s workloads.

Solution, Results and Benefits

As a Spot customer since 2016, Demandbase was successfully using Elastigroup by Spot to automate their EMR operations while reducing 70% of their EC2 cost.

So when Ocean by Spot was launched in 2018, it was a natural decision for the company to check it out.

While Spot is fully integrated with KOPS, Micheal decided to use EKS for most of their clusters, as EKS’s control plane is fully managed by AWS, eliminating unnecessary work by his team.

Just Set and Forget with Unparalleled EKS Workload Automation

For all their worker nodes, Ocean delivered a streamlined experience where Micheal and his team were able to get up and running in a few minutes.

“Using Ocean by Spot, it was simple to set up the clusters, choose the CPU and Memory and what’s amazing is that we have not needed to go back in and make any adjustments since,” said Micheal. He added “It’s much less labor overhead for us since Spot makes sure our clusters are immutable with our Pods always having the resources they need.”

Serverless Containers with Full Access to the Servers

With Ocean managing all the infra, but providing full access to the underlying EC2 instances, the Demandbase team could specify within which instance limitations they wanted to work with (e.g. families, specs, etc.), even allowing multiple family types and AZs, while delegating the actual sizing to Ocean.

Pod-Driven Autoscaling for Guaranteed Node Availability

Ocean’s Pod-driven autoscaling ensures that resource requests are filled in accordance with real-time requirements for Memory and CPU, resulting in Demandbase’s workloads always running as needed.

Automated Pod Rescheduling Simulations

Whenever more optimization is possible, Ocean stops Pod scheduling, proactively and gracefully drain any underutilized Nodes and bin-packs the containers onto the best suited instance(s).

Full Visibility into Cluster Details, Health and Efficiency

Micheal enthused “What I love about Ocean is its centralized, Kubernetes-native dashboarding where I can zoom into my cluster for a deep-dive into the nodes, namespaces, services and pods.”

This unprecedented visibility into the health and efficiency of Demandbase’s Kubernetes workloads was invaluable in keeping a strong understanding between the application developers and the infrastructure operators.

No More Manual Cluster Upgrades

For upgrading their control plane, EKS handled that with a simple click of the button. Likewise, for worker nodes that needing an update, Demandbase would provide the new AMI to Ocean, click on cluster roll and a Blue-green deployment would be automated, seamlessly handling all the Kubernetes requests.

Pod Right-Sizing and Intelligent Bin-Packing for Full Infra Utilization

Ocean aggregates all the Pod definitions and analyzes whether they are properly sized and the underlying infra is being efficiently utilized. This analysis, available in the Spot dashboard, enabled Micheal to see if Pod definitions were overprovisioned so he could right-size the Pod requirements.

Container-Native Showback

For the finance team at Demandbase, the Kubernetes-oriented cost-analysis reporting, allowed them to properly do cost-allocation on their multi-tenant clusters across teams and projects, with a spend breakdown based on Pods and not just Nodes.

70% Cost Optimization with Enterprise-Level SLA

In addition to all the container workload automation, Ocean ran Demandbase’s mission-critical clusters on spot instances, helping to optimize their EC2 cost by 70% while ensuring high availability.

World-Class Customer Engagement

With 24/7 customer support and dedicated solution architects, Demandbase received timely and expert assistance not only with Ocean, but with general Kubernetes issues.

In addition, many feature requests made their way into the Spot platform, highlighting how Spot is truly “customer obsessed”.

Micheal recounted, “I can’t count the number of times we’ve had an idea on a feature we’d like to see and then after talking with someone at Spot it became a reality.”