A relational database on a burstable EC2 instance shows periodic latency. It uses a 350 GB gp2 EBS volume. CloudWatch shows VolumeReadOps drops to under 10% of peak during the slow periods. What change should the SysOps admin make to achieve steady high I/O performance?

Migrating to a gp3 volume is the correct solution because gp3 volumes offer a baseline performance of 3,000 IOPS and 125 MiB/s, independent of volume size, and these can be provisioned higher. This ensures consistent high I/O performance, addressing the periodic latency caused by gp2's burst credit depletion. The 350 GB gp2 volume has a baseline of 1050 IOPS (3 IOPS/GB) and can burst to 3000 IOPS, but only for a limited time. When burst credits are exhausted, performance drops to the baseline, explaining the observed latency. Cold HDD (sc1) volumes are designed for throughput-oriented, infrequently accessed data and would worsen latency for a relational database. Switching to a memory-optimized instance family would not resolve I/O bottlenecks if the EBS volume is the limiting factor. Enabling unlimited burst mode on the EC2 instance affects network and CPU bursting, not EBS volume I/O performance.