Zipher

The spark-aware autoscaler is significantly less wasteful than the native autoscaler, as it takes into account the context of our job runs and avoids unnecessary scaling.

Zipher promised more than 35% savings on databricks and aws costs, we were skeptical but after a POC they delivered over 50% savings, and no adverse performance impacts

Zipher has consistently delivered measurable Databricks cost efficiencies while improving workload stability. It optimized cluster autoscaling by reducing unnecessary node churn and maintaining a steady balance between on-demand and spot capacity. We saw improved DBU utilization across interactive and job clusters, along with more predictable spend and faster job execution times which leads to reduce our oveall compute cost.