Music platform machine learning: Cloud technology modernization

The Challenge‍

As part of our client's migration to a Kubernetes (K8S) cluster, the client had reached its compute allocation limit with over 6,000 cores consumed. This created a bottleneck: there wasn’t enough capacity to migrate an additional 150 target pipelines, and job failures increased due to resource constraints.

What We Did‍

We performed in-depth analysis of reserved CPU limits versus actual CPU requests across the cluster. Based on those insights, we implemented targeted optimizations to align resource reservations more closely with actual usage.

The Result

• Reduced peak CPU limit by 25%.
• Cut average CPU usage by 30%, freeing up around 960 vCPUs for other pipelines.
• Reduced idle (unused) CPUs by 80%, eliminating wasteful, paid-for compute capacity.
• Delivered significant measurable financial savings through reduced compute costs and unlocked capacity for future workloads.