Examkin

Question

Difficulty: MediumHigh-Performing and Scalable Network Architectures

A financial modeling firm is deploying a tightly coupled High-Performance Computing (HPC) simulation workload on Amazon EC2 instances in a single AWS Region. The application relies on the Message Passing Interface (MPI) library for node-to-node communication. To achieve maximum performance, the network architecture must support low latency and inter-instance throughput of up to 100 Gbps100\text{ Gbps}. Which combination of network configuration and instance placement will meet these requirements?

  1. Launch the EC2 instances in a cluster placement group and configure them to use Elastic Fabric Adapter (EFA) interfaces.Answer
  2. B
    Launch the EC2 instances in a spread placement group and configure them to use standard Elastic Network Interfaces (ENIs).
  3. C
    Configure Amazon Route 53 latency-based routing to resolve internal instance hostnames across multiple Availability Zones.
  4. D
    Deploy the instances within a single subnet and configure stateful Network Access Control Lists (NACLs) to bypass security group evaluation.

Answer

Launch the EC2 instances in a cluster placement group and configure them to use Elastic Fabric Adapter (EFA) interfaces.
The correct answer is launching the instances in a cluster placement group and utilizing Elastic Fabric Adapter (EFA). A cluster placement group places instances logically close to each other within a single Availability Zone, enabling low-latency and high-throughput networking. EFA provides OS-bypass (bypassing the operating system kernel) which is necessary to achieve high throughput (up to 100 Gbps100\text{ Gbps}) and low latency for tightly coupled Message Passing Interface (MPI) applications.

Step-by-Step Solution

1
Identify the workload requirements.
Tightly coupled HPC simulation using MPI, requiring sub-millisecond node-to-node latency and up to 100 Gbps100\text{ Gbps} throughput.
Selecting the correct network interface and placement group requires matching the workload's specific performance profile.
2
Determine the appropriate network interface for MPI workloads.
Select Elastic Fabric Adapter (EFA) instead of a standard Elastic Network Interface (ENI).
EFA supports OS-bypass capabilities, allowing MPI applications to communicate directly with the network interface hardware, reducing latency and maximizing throughput.
3
Determine the optimal instance placement strategy.
Launch instances within a cluster placement group in a single Availability Zone.
Cluster placement groups ensure instances are physically close on the same network hardware fabric, minimizing latency and supporting high-throughput node-to-node communication.

Key Concept

Elastic Fabric Adapter (EFA) and Cluster Placement Groups for HPC Workloads
Estimated Time:1m 30s
Rate this question

Topics

High-Performing and Scalable Network Architectures

More questions from this topic