Basic information
| Field | Details |
|---|---|
| English / full name | High Performance Computing on AWS |
| Chinese | AWS 高性能计算架构 |
| Japanese | AWS の高性能コンピューティング |
| Review priority | 考试频率 ⭐⭐⭐ |
| Often confused with | AWS Batch / ParallelCluster / EFA / FSx for Lustre |
In one sentence
Identify whether compute, node communication, storage I/O, or data movement is the bottleneck before combining AWS components.
Core summary
Loosely coupled work fits Batch and elastic nodes; tightly coupled MPI emphasizes single-AZ cluster placement, EFA, and parallel file systems.
Study points
- ENA improves ordinary IP networking; EFA targets supported low-latency communication.
- FSx for Lustre serves parallel file I/O while S3 holds durable datasets and results.
- Batch manages jobs and queues; ParallelCluster manages HPC clusters.
- Spot needs retryable jobs or recoverable checkpoints.
Common pitfalls
- More instances are not automatically faster; communication or storage can be the bottleneck.
- A cluster placement group improves communication but does not provide Multi-AZ fault isolation.
Key memory
HPC is compute, low-latency communication, parallel storage, and scheduling.