These questions are independently created for study purposes. They are not official AWS exam questions, actual exam questions, leaked questions, or exam dumps.
SAA-048
For an HPC workload, you want to connect multiple EC2 instances with low-latency, high-bandwidth inter-node communication. Which setting related to instance placement is the most appropriate?
Choices
Explanation and answer
Read the explanation and correct answer (spoiler — try answering first)
Correct answer: A. Place the instances in a cluster placement group
Explanation
A cluster placement group packs instances physically close together within a single Availability Zone, achieving low latency and high network throughput between nodes. It is the placement strategy suited to HPC and tightly coupled distributed computing.
Per-choice explanations
A. Place the instances in a cluster placement group
Correct. Close placement minimizes inter-node communication latency and makes the most of the bandwidth between instances.
B. Place the instances in a partition placement group
Incorrect. A partition placement group isolates the blast radius per partition for large distributed systems such as HDFS and Cassandra. Low-latency communication is not its purpose.
C. Distribute the instances across multiple Availability Zones and manage them with an Auto Scaling group
Incorrect. Distributing across Availability Zones improves availability, but the physical distance between AZs increases communication latency. It is the opposite direction from the requirement.
D. Place the instances in a spread placement group
Incorrect. A spread placement group distributes instances onto separate physical hardware to avoid simultaneous failures. It improves availability but does not improve inter-node latency.
Practical Use
EC2 placement groups specify where instances are placed on the underlying hardware, and there are three strategies: cluster, spread, and partition. A cluster placement group packs instances physically close together within a single Availability Zone, minimizing inter-node network latency and maximizing bandwidth. Use it for workloads where inter-node communication is the performance bottleneck, such as HPC, distributed machine learning training, and tightly coupled parallel computing. A spread placement group does the opposite, placing each instance on a different rack (an independent unit of power and network) to avoid simultaneous failures, which suits protecting a small number of critical instances. A partition placement group divides instances into logical groups called partitions and ensures partitions do not share racks, which is used for failure isolation in large distributed data stores such as HDFS, Cassandra, and Kafka. As a design trade-off, a cluster placement group is confined to a single AZ precisely because it packs instances close together, concentrating them in the same failure domain and making it weak against AZ failures. If you need availability, you need a design such as preparing multiple cluster groups across AZs. Also, instances in a group are easier to place if you use the same instance type and launch them all at once; adding instances later can produce insufficient capacity errors. Additional factors for extracting network performance include using the Elastic Network Adapter (ENA) or the lower-latency Elastic Fabric Adapter (EFA), and enabling enhanced networking. On SAA-C03, terms such as "HPC," "low-latency inter-node communication," and "tightly coupled parallel processing" point to cluster; "distribute critical instances onto separate hardware" points to spread; and "failure isolation for a large distributed data store" points to partition. Not mixing up the correspondence between the three names and their purposes translates directly into points in this area.