Amazon EMR Serverless eliminates local storage provisioning, reducing data processing costs by up to 20%

This article announces serverless storage for Amazon EMR Serverless, eliminating local disk provisioning for Apache Spark workloads and reducing costs by up to 20%.

• Automatically handles shuffle operations without manual disk sizing
• Reduces data processing costs up to 20% with no storage charges
• Eliminates job failures from disk capacity constraints and data skew
• Enables independent compute and storage scaling for elastic resource usage
• Multi-tier aggregation architecture consolidates shuffle data efficiently
• Custom networking stack reduces mesh complexity and connection overhead
• Decouples storage from compute, allowing instant worker release
• Available in multiple AWS regions for EMR release 7.12 and later
• Data encrypted in transit and at rest with job-level isolation
• Best for workloads shuffling 10GB-200GB with unpredictable sizes

Serverless storage shifts data processing infrastructure management by automating shuffle operations, aligning costs to actual usage, and improving reliability for I/O-intensive analytics workloads.