Q20 — AWS SAA-C03 Ch.17

Question 20 of 89 | ← Chapter 17

Q1320. A company is deploying an application that processes large quantities of data in parallel. The company plans to use Amazon EC2 instances for the workload. The network architecture must be configurable to provide the lowest possible latency between nodes.Which combination of network solutions will meet these requirements?(Select TWO.)

Correct Answer: B. Attach an Elastic Fabric Adapter(EFA) to each EC2 instance., E. Run the EC2 instances in a cluster placement group.

Explanation

The correct combination of network solutions to meet the requirements of low-latency communication between EC2 instances processing large data in parallel is:Correct Answers:B. Attach an Elastic Fabric Adapter (EFA) to each EC2 instance.E. Run the EC2 instances in a cluster placement group.Why These Options Are Correct:B. Elastic Fabric Adapter (EFA)EFA is a network interface that enhances inter-node communication for High-Performance Computing (HPC) and Machine Learning (ML) workloads.It provides OS-bypass capabilities, allowing direct access to the network interface from applications (via libraries like MPI, OpenMPI, or NCCL), reducing latency and improving throughput. Best for: Low-latency, high-bandwidth workloads (e.g., HPC, distributed training, financial simulations).E. Cluster Placement GroupA placement group is a logical grouping of EC2 instances within a single Availability Zone (AZ) that minimizes network latency and maximizes throughput between instances. Cluster placement groups are recommended for tightly coupled, latency-sensitive workloads (e.g., HPC, distributed databases, real-time analytics).Key benefit: Instances are placed in close physical proximity, reducing network hops and improving communication speed.Why Other Options Are Incorrect:A. Distribute the EC2 instances across multiple Availability Zones Incorrect because cross-AZ communication introduces higher latency due to inter-AZ network traffic. Best for: High availability (HA) and fault tolerance, not low-latency workloads.C. Place the EC2 instances in a single Availability ZonePartially correct (low-latency within an AZ), but not sufficient alone because it doesn't optimize instance placement for minimal latency.Cluster placement groups (E) are needed for optimal low-latency networking within an AZ.D. Use Amazon EBS-optimized instance typesIncorrect because EBS optimization improves storage I/O performance, not inter-node network latency. Best for: Workloads requiring high disk throughput (e.g., databases, big data processing), not low-latency networking.Best Practice Combination:Use a Cluster Placement Group (E) to ensure instances are physically close. Attach EFAs (B) to enable OS-bypass for ultra-low-latency communication. (Optional) Use EBS-optimized instances (D) if storage performance is also critical, but this does not affect networking latency.Example Architecture:Workload: Distributed HPC or ML training job.EC2 Instances:Instance Type: p4d.24xlarge (supports EFA).Placement Group: Cluster (single AZ).Networking: EFA-enabled for MPI/NCCL communication.Result: Sub-millisecond latency between nodes, maximizing parallel processing efficiency.Conclusion:The best choices for low-latency parallel processing are:B (EFA) + E (Cluster Placement Group).Other options either increase latency (A) or do not improve networking (C, D).