Examkin

Question

Difficulty: MediumAuto-scaling and Capacity Planning

An architectural modeling platform processes asynchronous 3D spatial rendering tasks submitted by enterprise users. The processing worker nodes are deployed on Compute Engine Managed Instance Groups (MIGs). Tasks are pulled from a Cloud Pub/Sub topic and processed by worker VMs, which spend most of their processing time waiting on I/O operations and database queries. The platform expects a fivefold traffic burst during an upcoming product launch. Which TWO actions should a Cloud Architect recommend to ensure the rendering system auto-scales efficiently and meets capacity demands during the burst? (Select TWO.)

  1. Configure the Managed Instance Group auto-scaling policy to scale based on a Cloud Monitoring custom metric tracking unacknowledged messages per worker instance.Answer
  2. Audit regional Compute Engine vCPU quota limits in advance and submit quota increase requests for the target regions prior to the launch.Answer
  3. C
    Set the Managed Instance Group auto-scaling metric strictly to target an average CPU utilization of 80% across all instances.
  4. D
    Migrate the processing worker application to Google Kubernetes Engine (GKE) Autopilot to eliminate the need for capacity planning and compute quota management.
  5. E
    Purchase 3-year Committed Use Discounts (CUDs) for peak burst capacity to bypass regional vCPU quota checks during scaling events.

Answer

The platform should scale the Managed Instance Group based on a Cloud Monitoring custom metric tracking unacknowledged messages per worker instance, and proactively request regional Compute Engine vCPU quota increases prior to the launch event.
For I/O-bound queue processing workloads, auto-scaling should be based on task backlog metrics (such as Cloud Pub/Sub unacknowledged message count per instance) rather than CPU usage. Additionally, capacity planning requires verifying and requesting vCPU quota increases ahead of anticipated traffic bursts so that auto-scaling is not blocked by platform quota limits.

Step-by-Step Solution

1
Analyze workload characteristics
Identify that the workload is I/O-bound and queue-driven by Cloud Pub/Sub.
CPU utilization does not accurately reflect work backlogs for I/O-bound worker processes.
2
Select appropriate auto-scaling metric
Choose queue backlog metrics (unacknowledged messages per instance) via Cloud Monitoring.
This guarantees that the MIG scales out when task backlog increases regardless of CPU load.
3
Perform capacity planning
Check regional Compute Engine vCPU quotas and request increases before the event.
MIG auto-scalers are constrained by project quota limits and cannot spawn instances if quotas are exhausted.

Key Concept

Auto-scaling metric selection for queue-driven/I-O bound workloads and proactive vCPU quota management in GCP.
Rate this question

Topics

Auto-scaling and Capacity Planning

More questions from this topic