Question

Difficulty: MediumPlanning Google Kubernetes Engine (GKE) Cluster Architectures

A research institute is designing a Google Kubernetes Engine (GKE) cluster architecture to execute large-scale, fault-tolerant batch analysis jobs. The workloads require specialized host-level OS kernel modifications via custom `sysctl` settings on the underlying node OS, and the organization wants to minimize compute infrastructure expenses as much as possible. Which GKE cluster architecture should the cloud engineer select?

  1. Provision a GKE Standard cluster with a dedicated node pool configured to use Spot VMs.Answer
  2. B
    Provision a GKE Autopilot cluster utilizing Spot pods for all batch workloads.
  3. C
    Provision a GKE Standard cluster with On-Demand VM node pools while assigning Spot VMs to the cluster control plane.
  4. D
    Provision a GKE Autopilot cluster using On-Demand pods configured with Horizontal Pod Autoscaler (HPA) to scale worker node capacity.

Answer

Provision a GKE Standard cluster with a dedicated node pool configured to use Spot VMs.
GKE Standard mode provides administrative access and flexibility needed to customize node OS configurations, such as custom sysctl settings. Combining GKE Standard with Spot VMs in node pools satisfies the cost minimization requirement for fault-tolerant workloads.

Step-by-Step Solution

1
Evaluate operational mode requirements (Autopilot vs. Standard)
GKE Standard must be selected because the workload requires custom node OS kernel modifications (sysctl settings), which are restricted in GKE Autopilot.
GKE Autopilot locks down the underlying host nodes to ensure security and automated management, prohibiting host-level custom OS modifications.
2
Evaluate compute cost optimization options for fault-tolerant workloads
Configure the node pool to use Spot VMs.
Spot VMs provide up to 60-91% discounts compared to standard On-Demand instances, making them ideal for fault-tolerant, stateless batch processing jobs.

Key Concept

Selecting between GKE Standard and Autopilot based on node OS customization constraints and leveraging Spot VMs for cost-effective batch workloads.
Rate this question