Examkin

Soru

Zorluk: ZorHigh-Performing and Scalable Storage Solutions

An engineering firm is deploying a seismic simulation workload on AWS. The workload consists of a cluster of 5050 Amazon EC2 Linux instances that require concurrent, parallel read and write access to a shared file system. The simulation demands sub-millisecond latencies, an aggregate throughput of at least 15 GiB/s15\text{ GiB/s}, and over 100000 IOPS100{}000\text{ IOPS}. The data is highly transient scratch data that can be easily regenerated if lost. Which storage configuration meets these requirements with the highest performance and the lowest operational complexity?

  1. A
    An Amazon Elastic File System (Amazon EFS) file system configured with Max I/O performance mode and Provisioned Throughput set to 15 GiB/s15\text{ GiB/s}.
  2. B
    An Amazon EBS io2 Block Express volume configured with Multi-Attach enabled and formatted with a clustered file system.
  3. An Amazon FSx for Lustre file system using SSD storage deployed in scratch mode.Cevap
  4. D
    An Amazon EBS gp3 volume formatted with an ext4 file system and attached directly to all EC2 instances.

Cevap

An Amazon FSx for Lustre file system using SSD storage deployed in scratch mode
The correct option is the FSx for Lustre deployment because it is a native parallel file system designed specifically for high-performance computing (HPC) and scratch workloads. It supports parallel access from hundreds of EC2 instances and scales performance (both throughput and IOPS) with SSD storage while maintaining sub-millisecond latencies.

Adım Adım Çözüm

1
Analyze the access requirement
The architecture requires parallel, concurrent read and write access from a cluster of 5050 Linux EC2 instances.
This rules out standard Amazon EBS volumes which cannot be attached to multiple instances concurrently, and EBS Multi-Attach which is limited to a maximum of 1616 instances.
2
Evaluate throughput and latency constraints
The file system must support sub-millisecond latencies, at least 15 GiB/s15\text{ GiB/s} throughput, and over 100000 IOPS100{}000\text{ IOPS} for transient scratch data.
Amazon EFS cannot scale to 15 GiB/s15\text{ GiB/s} write throughput, and its Max I/O mode increases latency for metadata operations, making it unsuitable.
3
Match workload characteristics to optimal service
Amazon FSx for Lustre is a high-performance parallel file system designed for HPC workloads, supporting parallel client access, sub-millisecond latency, and massive scaling.
Deploying it as SSD scratch storage matches the transient nature of the data and provides the maximum cost-effective performance.

Anahtar Kavram

High-Performance Parallel Storage for HPC Workloads
Bu soruyu puanla

Konular

High-Performing and Scalable Storage Solutions

Bu konudan daha fazla soru