Examkin

Soru

Zorluk: OrtaAuto-scaling and Capacity Planning

An IoT enterprise telemetry platform processes streaming sensor data using Compute Engine Managed Instance Groups (MIGs). The application is network I/O-bound, keeping thousands of concurrent TCP sockets open per instance. During peak ingestion periods, telemetry drops occur due to socket exhaustion while average CPU utilization remains at approximately 30%. Furthermore, during a recent sudden traffic surge, manual scale-out attempts failed because the project exceeded its regional vCPU quota limit. Which combination of actions should a cloud architect implement to ensure automatic, reliable scaling and capacity availability during peak loads?

  1. Configure the MIG autoscaler to scale using a custom Cloud Monitoring metric tracking active TCP connection count, and proactively request a regional Compute Engine vCPU quota increase for the instance family.Cevap
  2. B
    Lower the target CPU utilization threshold on the existing MIG autoscaler policy from 80% to 25% to trigger earlier scaling, while keeping current project quotas.
  3. C
    Configure the MIG autoscaler using a custom Cloud Monitoring metric for active connections, but rely on Google Cloud's default auto-quota expansion during scale-out events.
  4. D
    Migrate the telemetry ingestion service to a multi-zone Google Kubernetes Engine (GKE) cluster with Cluster Autoscaler to automatically bypass Compute Engine regional vCPU quotas.

Cevap

Configure the MIG autoscaler using a custom Cloud Monitoring metric tracking active TCP connection count, and proactively request a regional Compute Engine vCPU quota increase.
For network I/O-bound workloads with low CPU usage, autoscaling must be based on custom application metrics such as open TCP connections or socket usage. Additionally, because compute instances cannot scale past project quotas, requesting regional vCPU quota increases in advance ensures capacity is available when autoscaling triggers.

Adım Adım Çözüm

1
Identify the performance bottleneck metric
Recognize that CPU utilization is an inadequate metric for I/O-bound socket workloads.
I/O-bound services exhaust memory or network sockets long before CPU thresholds are reached.
2
Select the appropriate autoscaling metric strategy
Implement custom metrics via Cloud Monitoring for active TCP connections.
Custom metrics reflect real load and trigger scale-out events before socket exhaustion occurs.
3
Address capacity provisioning constraints
Request regional Compute Engine vCPU quota increases in advance.
Autoscaling policies cannot provision new instances if regional resource quota limits are exceeded.

Anahtar Kavram

Custom Metric Autoscaling & Preemptive Quota Management
Bu soruyu puanla

Konular

Auto-scaling and Capacity Planning

Bu konudan daha fazla soru