Optimizing Price and Performance with Enhanced Amazon EBS GP3 Volumes

Decoupling Performance and Capacity in EBS

• EBS allows customers to decouple the performance and capacity of their storage volumes
• This is in contrast to traditional storage systems where performance and capacity are tightly coupled
• With GP3 volumes, customers can independently scale IOPS and throughput without having to overprovision capacity
• This enables more cost-effective optimization based on specific workload requirements

Coupled vs. Decoupled Storage

• GP2 volumes provide coupled performance and capacity, similar to physical disks
• GP3 volumes offer decoupled performance and capacity, allowing independent scaling
• Decoupled storage enables building a volume strategy tailored to specific workload needs
  • Transactional workloads can optimize for IOPS
  • Throughput-heavy workloads can optimize for throughput

Monitoring and Managing EBS Performance

Latency, IOPS, and Queue Depth

• Latency, IOPS, and queue depth (QD) are fundamentally connected by queuing theory
• Increasing queue depth can increase IOPS up to the volume's IOPS entitlement
• Beyond the IOPS entitlement, increasing queue depth will increase latency
• Customers can use this relationship to control and optimize latency for their workloads

Latency Histograms

• Latency histograms provide a detailed view of an EBS volume's latency profile
• Histograms show the distribution of latency across I/O operations
• This allows identifying typical latency as well as outliers and variability
• Tracking latency histograms over time can reveal performance changes and issues

Handling Latency Outliers

• Latency outliers can occur due to physical SSD failures and block remapping
• EBS proactively monitors for and mitigates these outliers to maintain the expected latency SLA
• Techniques used include:
  • Routing I/O away from misbehaving servers
  • Read and write hedging to avoid slow responses
  • Fault injection testing to understand application resilience

Evolving EBS Volumes Over Time

Modifying EBS Volumes with EBS Elastic Volumes (EV)

• EV allows dynamically increasing size, IOPS, and throughput of EBS volumes
• This enables customers to right-size their volumes as workload requirements change
• Key EV capabilities:
  • Migrate between volume types (e.g., GP2 to GP3)
  • Scale IOPS and throughput within the GP3 range
  • Increase volume size (but not decrease)

Optimizing Secondary Environments

• For dev/test environments, customers can leverage GP3 to optimize cost vs. higher-performance options like IO2
• EBS snapshots are optimized to only store incremental changes, reducing storage costs
• EBS clones provide instant volume copies without the need for restoring from snapshots

Key Takeaways

• EBS GP3 volumes decouple performance and capacity, enabling cost-effective optimization
• Monitoring latency, IOPS, and queue depth is crucial for understanding and tuning EBS performance
• EBS Elastic Volumes allow dynamically evolving volumes to match changing workload requirements
• Customers can optimize costs in secondary environments by leveraging GP3 and EBS snapshot/clone capabilities