Kubernetes Cost Optimization: From Roadblocks to Results

Trax deployed PerfectScale by DoiT to gain AI-guided cost visibility and optimization recommendations across their 200+ microservice Kubernetes environment. The platform delivered clear visibility into resource needs per service and surfaced the biggest waste-elimination opportunities, while balancing cost savings against resilience. PerfectScale's unique ability to consolidate every replica of a service into a single view enabled Trax to understand heterogeneous utilization across replicas — particularly valuable for ephemeral workloads like Spark and Flink jobs — and rearchitect services for further cost savings without impacting availability.

Trax's mission is to enable brands and retailers to harness the power of digital technologies to produce the best shopping experiences imaginable. Their industry-leading innovations and excellence through the development of advanced technologies and autonomous data collection methods drive positive shopper experiences and unlock revenue opportunities at all points of sale. Trax's solution portfolio provides mission-critical metrics, analytics, and services that help customers save time and money by improving their shopping experience. Kubernetes is a key component of their infrastructure, supporting a large-scale multi-cloud, multi-cluster environment serving customers in over 90 countries, including some of the globe's largest enterprises.

At the beginning of the year, the CFO laid out aggressive cost savings goals throughout multiple aspects of the organization. For Mark Serdze, Director of Cloud Infrastructure, and his team, this meant quickly taking action to optimize cloud cost. They achieved fast results outside of Kubernetes, but inside their clusters they hit roadblocks. 'We started optimizing manually with available metrics, using Vertical Pod Autoscaler (VPA), cluster logs, and our monitoring solutions,' explained Serdze. 'This approach didn't give us proper clarity and would be challenging to scale efficiently without big development needs. This left us taking ad-hoc, reactive actions that were having minimal impact on our goals.' Gaps in their toolset created time-consuming debate over which actions to take, often resulting in hunch-based decisions, additional rework, and risk to service resiliency.

Shortly after deploying PerfectScale by DoiT, Serdze and the team gained the cost visibility they were missing across their Kubernetes environment. Across their vast 200-plus microservice environment, they had clear visibility into what resources each service needed and where the biggest waste-elimination opportunities existed. The platform's AI-guided intelligence allowed them to quickly take action, comparing cost savings with overall resilience to adjust resources safely and efficiently without compromising performance. 'The cost optimization recommendations were key for us, telling us what actions to take with a clear understanding of the impact each change would have,' Serdze explained. 'In one of our clusters, we were able to reduce cost by 75%, saving us over 6-figures in yearly expenses.' The comprehensive data and intelligence also let Trax replace an existing FinOps tool that lacked granular cost details and optimization guidance — with virtually no impact on their budget.

After eliminating wasted resources, the Trax team turned to identifying additional cost optimization opportunities tied to business outcomes. 'A key metric for us is cost per processing, which is heavily affected by our Kubernetes efficiency,' said Serdze. 'If it gets over a certain amount, we are under a lot of pressure to figure out why and to take actions to reduce it.' PerfectScale's unique feature that consolidates every replica of a service into a single view gave the team clarity on heterogeneous utilization across replicas — particularly useful for ephemeral workloads like Spark or Flink jobs. 'We were able to build multiple flavors of the service with different levels of resources and route the incoming requests to the proper service based on the size of the data,' explained Serdze. 'This made a big impact on our cost per processing metric. PerfectScale surfaced this data instantly, and without them, we would have spent countless hours evaluating hundreds of replicas to generate the same results.'

Beyond the quick and effective cost-optimization results, the Trax team attributed much of their success to the strategic partnership and support they received from PerfectScale. 'The support during the proof of concept (POC) made a big impact in helping us drive quick results,' said Serdze. 'The PerfectScale team sat with us, helped us optimize, and ensured our success in using the platform. I have not seen this level of commitment from other vendors, and I am glad we found a partner we can rely on to help us keep our Kubernetes cost in check as we continue to scale.'