Examkin

Soru

Zorluk: Çok zorSelecting and Designing Compute and Application Platforms

A genomics research institute is architecting a cloud-native pipeline on Google Cloud to process high-throughput DNA sequencing data. The architecture comprises two distinct workload requirements:

1. A high-performance computing (HPC) sequence alignment engine requiring custom Linux kernel drivers, message passing interface (MPI) inter-node communications with sub-millisecond network latency, and maximum I/O performance on local NVMe storage.
2. An edge-facing containerized API layer that receives large binary sequence data via continuous gRPC streams with individual session durations lasting up to 45 minutes, experiences unpredictable bursts, and must scale to zero instances during idle periods with zero infrastructure management overhead.

Which TWO compute platform choices and configurations should you implement to satisfy both workload requirements while maintaining architectural best practices? (Select TWO.)

  1. Provision Compute Engine virtual machines equipped with local NVMe SSDs, Tier 1 networking, gNIC interfaces, and configured within a compact placement policy for the sequence alignment engine.Cevap
  2. Deploy the containerized web API layer on Cloud Run configured with HTTP/2 enabled, session affinity, and an extended request execution timeout of up to 60 minutes.Cevap
  3. C
    Deploy the sequence alignment engine on Cloud Run using Serverless VPC Access and mounting Cloud Storage FUSE buckets for shared file storage.
  4. D
    Deploy the containerized gRPC streaming API on Google Kubernetes Engine (GKE) Autopilot using a custom DaemonSet to orchestrate node scale-to-zero operations.
  5. E
    Deploy both the sequence alignment engine and the gRPC API layer onto App Engine Flexible environment using custom Docker runtime images.

Cevap

The optimal architecture combines Compute Engine virtual machines configured with compact placement policies and local NVMe SSDs for the HPC alignment engine, alongside Cloud Run with HTTP/2 and extended timeouts for the containerized gRPC streaming API.
For specialized HPC workloads requiring custom Linux kernel modules, sub-millisecond MPI latency, and high-performance disk access, Compute Engine with compact placement policies, gNIC, and local NVMe SSDs is the only GCP compute platform that satisfies hardware and OS control constraints. For stateless or streaming containerized APIs requiring long timeouts (up to 60 minutes) and scale-to-zero functionality, Cloud Run provides full gRPC streaming support with zero server management overhead.

Adım Adım Çözüm

1
Analyze the HPC sequence alignment workload requirements
Identified requirements for custom OS kernel modules, low-latency inter-node MPI communication, and raw local I/O throughput.
Serverless and managed container platforms restrict host kernel access and do not support compact placement policies. Compute Engine VMs with compact placement policies and gNIC are required.
2
Analyze the containerized gRPC API workload requirements
Identified requirements for long-running gRPC streaming (up to 45 minutes), scale-to-zero capability, and minimal operational overhead.
Cloud Run supports containerized workloads, native HTTP/2 gRPC streaming, request timeouts up to 60 minutes, and automatic scale-to-zero without cluster management overhead.
3
Evaluate and reject misallocated compute platforms
Rejected GKE, App Engine Flexible, and Cloud Run for the HPC engine due to missing kernel/placement capabilities; rejected GKE for the API to avoid management overhead.
Matching workload operational profiles to the appropriate GCP compute abstraction optimizes performance while reducing baseline costs and operational complexity.

Anahtar Kavram

Compute Platform Selection for Specialized HPC vs. Serverless Container Workloads
Bu soruyu puanla

Konular

Selecting and Designing Compute and Application Platforms