AWS Certified Solutions Architect – ProfessionalAccelerate Workload Migration and ModernizationHard
A research institution is migrating a high-performance computing (HPC) cluster to AWS. The cluster consists of tightly coupled applications that require ultra-low latency networking (sub-millisecond) and high inter-node bandwidth for inter-process communication (MPI). The existing on-premises cluster uses InfiniBand interconnects. Which AWS EC2 instance type and networking feature combination BEST meets these demanding requirements?
- AC7g instances with Elastic Network Adapters (ENA)
- BR6gn instances with ENA Express
- CM6i instances with Enhanced Networking (Intel 82599 Virtual Function)
- DHpc6a instances with Elastic Fabric Adapter (EFA)
Show answer & explanationAnswer & explanation
Correct answer: D. Hpc6a instances with Elastic Fabric Adapter (EFA)
Hpc6a instances are specifically designed for HPC workloads, offering high core counts and memory. The Elastic Fabric Adapter (EFA) is a network interface that enables HPC applications to achieve lower and more consistent latency and higher throughput than traditional TCP networking, similar to on-premises HPC clusters with InfiniBand, making it ideal for tightly coupled MPI workloads.
Why the other options are wrong
- A. C7g instances are general-purpose compute-optimized (Graviton3) and ENA provides good networking, but not the ultra-low latency and high bandwidth required for tightly coupled HPC with MPI.
- B. R6gn instances are memory-optimized (Graviton2) with good networking (ENA Express for higher bandwidth), but EFA is the critical component for the specific inter-process communication requirements of tightly coupled HPC.
- C. M6i instances are general-purpose and Enhanced Networking (VF) is an older term for improved networking, not specific to the ultra-low latency and high bandwidth of HPC via EFA.
Elastic Fabric Adapter (EFA)
A network interface for Amazon EC2 instances that enables customers to run applications requiring high levels of inter-node communications at scale on AWS, like HPC and machine learning.
- Provides lower and more consistent latency than traditional TCP.
- Supports OS-bypass capabilities for MPI and NCCL.
- Similar performance to on-premises InfiniBand networks.
Memory trick: EFA gives your HPC apps 'Extra Fast' networking, like InfiniBand.