SAA-C03 Practice Question 38: Cluster Placement Group Latency
AWS Certified Solutions Architect – Associate (SAA-C03) · Test 4 · question 8 of 10
A tightly coupled HPC simulation runs across 20 EC2 instances that exchange messages continuously. Profiling shows the job is bound by node-to-node network latency rather than by CPU. The team wants the lowest possible latency and highest packet-per-second performance between the instances, and accepts that all 20 running in one Availability Zone is an acceptable risk for a job that is restarted on failure.
What should the solutions architect recommend?
Show answer and explanation
Answer:
A cluster placement group packs instances close together on the same high-bisection-bandwidth network segment within one Availability Zone, which is the configuration AWS provides for tightly coupled HPC traffic. Combined with enhanced networking it delivers the lowest inter-instance latency and the highest packet-per-second rate available. The stem explicitly accepts the single-AZ concentration that this entails, which is the trade-off a cluster placement group makes.
Option B optimises for the opposite goal. A spread placement group deliberately places instances on distinct underlying hardware to minimise correlated failure, which increases the distance between them and therefore the latency.
Option C also targets fault isolation. Partition placement groups separate instances into partitions that do not share racks, and are aimed at large distributed data stores such as HDFS or Cassandra, not at minimising latency between every pair of nodes.
Option D adds a network hop to traffic that is bound by latency, and spreading across Availability Zones introduces inter-AZ latency on every message. A load balancer is also the wrong shape for peer-to-peer message passing between compute nodes.
Reference: Placement groups for Amazon EC2 instances