GCP Professional Cloud Architect Practice Question

Your company runs a nightly Monte-Carlo risk-analysis job that needs to burst to about 40 000 vCPUs for 3-4 hours in europe-west1. The batch application checkpoints every five minutes and can restart on a different VM if one is lost, so occasional interruptions are acceptable. During the same time window, a customer-facing web tier in the same region must always have 32 vCPUs available and cannot tolerate preemption. You must minimize total compute cost while ensuring that both workloads finish on schedule. Which approach best meets these requirements?

Deploy the web tier on sole-tenant nodes for guaranteed resources and run the simulation on standard, on-demand VMs that scale out with an instance-based autoscaler.
Run both workloads on standard on-demand VMs covered by a three-year committed-use discount and place them in an unmanaged instance group.
Place the simulation in a regional managed instance group that uses Spot VMs; create a Compute Engine capacity reservation for 32 standard VMs to host the web tier during the burst window.
Launch both workloads on preemptible VMs in an autoscaled managed instance group, assigning the web tier a higher priority so its instances are less likely to be preempted.

Report Issue

Answer Description

The bursty, fault-tolerant simulation can run on a managed instance group that uses Spot VMs, which cost up to 91 % less than standard VMs but may be preempted at any time. Because the job checkpoints frequently, it can absorb those interruptions without extending overall runtime, giving the lowest possible cost. The latency-sensitive web tier needs guaranteed capacity and cannot be preempted, so creating a Compute Engine capacity reservation for 32 standard VCPU-equivalent instances ensures they are always available at on-demand pricing without the overhead of dedicated (sole-tenant) nodes. Committed use discounts do not guarantee capacity, and relying on autoscaler priorities cannot prevent Spot or preemptible VMs from being preempted. Therefore, using Spot VMs in a managed instance group for the simulation together with a capacity reservation of standard VMs for the web tier is the most cost-effective and reliable solution.

Ask Bash

Bash is our AI bot, trained to help you pass your exam. AI Generated Content may display inaccurate information, always double-check anything important.