Accelerate Spark on EMR Serverless with larger workers and shuffle-optimized disks

This article announces new 32 vCPU / 244 GB worker configurations on Amazon EMR Serverless with shuffle-optimized disk support for demanding analytics workloads.

• New 32 vCPU workers support up to 2,000 GB shuffle-optimized disk versus 200 GB standard disk on smaller workers
• Benchmark results across 126 TPC-DS and TPC-H queries show 29% average performance improvement and 29% lower costs
• Shuffle-heavy queries achieved 45–55% improvements; shuffle-optimized disk provides higher IOPS and throughput for shuffle operations
• Large workers benefit shuffle-intensive, I/O-heavy, and memory-intensive workloads by keeping shuffle data local and reducing remote fetches
• Configuration provided: 6 executors × 32 vCPU with 220 GB memory and 2,000 GB shuffle-optimized disk

EMR Serverless larger workers enable migration of heavyweight Spark workloads while maintaining serverless simplicity and improving price-performance.