Design High-Performing Architectures

A self-managed transactional database needs consistent, very high random IOPS with low latency on its EBS volume. Which EBS volume type best fits?

• AThroughput Optimized HDD (st1), designed for large sequential workloads.
• BCold HDD (sc1), designed for infrequently accessed cold data storage.
• CProvisioned IOPS SSD (io2), designed for high, consistent random IOPS.
• DA general-purpose gp2 volume left at its default baseline performance.

Provisioned IOPS SSD (io2) delivers high, consistent random IOPS with low latency for demanding transactional databases. HDD types (st1, sc1) are for sequential or cold data, and a default gp2 volume may not sustain the required IOPS.

A machine learning training cluster needs a shared file system offering very high throughput and low latency, integrated with data in Amazon S3, for high-performance computing. Which service fits best?

• AAmazon FSx for Lustre, a high-performance file system integrated with S3.
• BAmazon EFS Standard, a general-purpose shared file system for Linux.
• CAn Amazon S3 bucket accessed directly through the standard object API.
• DA single large EBS volume attached to one instance in the cluster.

FSx for Lustre provides a high-throughput, low-latency parallel file system for HPC and ML, with native S3 integration. EFS is general-purpose, direct S3 object access lacks a file-system interface and Lustre performance, and a single EBS volume cannot be shared across the cluster.