AWS Certified Solutions Architect – ProfessionalAccelerate Workload Migration and ModernizationHard
A research institution is migrating a high-performance computing (HPC) cluster to AWS. This cluster runs tightly coupled scientific simulations that require extremely low-latency, high-throughput network communication between compute nodes. The institution wants to ensure the migrated workload achieves comparable performance to its on-premises InfiniBand network. Which AWS networking feature is MOST critical to meet these requirements?
- AElastic Fabric Adapter (EFA)
- BAWS Direct Connect for dedicated connectivity
- CEnhanced Networking with Elastic Network Adapter (ENA)
- DJumbo Frames for increased Maximum Transmission Unit (MTU)
Show answer & explanationAnswer & explanation
Correct answer: A. Elastic Fabric Adapter (EFA)
Elastic Fabric Adapter (EFA) is a network interface for Amazon EC2 instances that enables customers to run HPC applications requiring high levels of inter-node communication at scale on AWS. It provides lower latency and higher throughput compared to traditional TCP networking, similar to on-premises InfiniBand.
Why the other options are wrong
- B. AWS Direct Connect provides a dedicated network connection from on-premises to AWS, improving overall connectivity but not the inter-node communication latency and throughput *within* the AWS HPC cluster itself.
- C. ENA provides improved networking performance over standard EC2 networking but does not offer the specialized, low-latency, kernel-bypass capabilities required for tightly coupled HPC workloads like EFA.
- D. Jumbo Frames (MTU up to 9001) can increase throughput by reducing packet overhead but do not address the fundamental latency and inter-node communication architecture needs of tightly coupled HPC applications.
Elastic Fabric Adapter (EFA)
A network interface for Amazon EC2 instances that enables tightly coupled HPC applications to achieve lower latency and higher throughput, similar to on-premises InfiniBand.
- Provides kernel-bypass networking and OS-bypass capabilities.
- Optimized for inter-node communication in HPC and ML.
- Reduces latency and increases throughput significantly.
Memory trick: EFA for Extreme Fast HPC