Run Spark 31% faster and optimize compute costs with Amazon S3 Express One Zone on Amazon EMR

This article demonstrates how Amazon S3 Express One Zone improves Apache Spark performance on Amazon EMR through benchmarking with TPC-DS datasets.

• S3 Express One Zone reduced total benchmark runtime by 31% at 3TB scale and 47% at 10TB scale on an 8-node Graviton4 cluster
• 103 of 105 TPC-DS queries ran faster with no application code changes required
• Per-request latency dropped from 121ms to 7ms, a 17-fold reduction in read latency
• Overall cost per benchmark run decreased 36% due to shorter compute time and 85% lower S3 request pricing
• Recommended pattern: write data to both S3 Express One Zone and S3 Standard, use lifecycle rules to expire hot tier data after retention window
• Most I/O-heavy queries saw largest gains, with query q76 achieving 62.7% speedup
• Setup requires co-locating EMR cluster in same Availability Zone as directory bucket and disabling fast partition discovery

S3 Express One Zone delivers significant performance and cost benefits for I/O-bound analytics workloads, with benefits increasing as data volume grows.