AWS re:Invent 2025 - Maximizing block storage performance for high-intensity workloads (STG319)

Maximizing Block Storage Performance for High-Intensity Workloads

Overview

• Presentation by Mark Olsen, Senior Principal Engineer on the EBS team, and Jody, EBS Product Manager
• Focuses on optimizing Amazon Elastic Block Store (EBS) performance for mission-critical applications
• Discusses planning, testing, and monitoring strategies to ensure consistent, high-performance storage

Planning Your Infrastructure

• Understand your application's workload requirements
  • For a transactional database, use high-performance IO2 volumes with 256K IOPS and 59-9s durability
  • Leverage different volume types (IO2, GP3) for different components (e.g. database journals, Kafka topics)
• Leverage larger, faster GP3 volumes
  • Increased max IOPS from 16K to 80K, throughput from 1,000 to 2,000 MB/s, and max size from 16 TB to 64 TB
• Understand performance characteristics of volume types
  • GP3: Designed for single-digit millisecond latencies 99% of the time
  • IO2: Designed for sub-millisecond latencies 99.9% of the time, with 10x fewer latency outliers than GP3

Testing and Benchmarking

• Use flexible tools like FIO for raw I/O testing
• Leverage industry-standard benchmarks like TPC-C to simulate realistic workloads
• Perform end-to-end load testing on your application to uncover bottlenecks
• Use AWS Fault Injection Service (FIS) to test resilience under controlled chaos
  • New EBS-specific actions to inject latency and simulate degraded performance

Monitoring and Troubleshooting

• Analyze instance status checks to identify infrastructure vs. application issues
  • Check for problems with the instance, EBS attachment, or instance liveness
• Leverage new EBS performance metrics for deeper visibility
  • Average IOPS, throughput, and latency histograms per volume
• Understand how burst capacity works on instance types like M5 and how to size instances appropriately
• Use elastic volumes to change volume type, size, and performance characteristics on the fly
  • Caveats around latency impact during optimization

The Evolution of EBS Performance

• EBS has continuously improved in performance and capabilities over the years
• Nitro-based instances provide hardware offloads and a custom SRD network protocol for lower latency and higher throughput
• New R5b instances offer up to 720K IOPS and 150 Gbps of EBS bandwidth

Business Impact

• Consistent, high-performance storage is critical for mission-critical applications like medical device manufacturing and healthcare software
• Proper planning, testing, and monitoring can help ensure resilience, compliance, and optimal performance
• Leveraging the latest EBS features and instance types can provide a significant boost in throughput and IOPS for demanding workloads

Key Takeaways

• Understand your application's storage requirements and choose the right EBS volume types
• Thoroughly test your application's performance and resilience using tools like FIO, industry benchmarks, and AWS FIS
• Monitor EBS performance metrics closely and leverage elastic volumes to adapt to changing needs
• Take advantage of the latest EBS and EC2 instance advancements for maximum performance