Knowledge Check - Compute and Elasticity
Test your understanding of the key concepts from Module 4. Select the best answer for each question, then submit to see your results. You need to score at least 60% to pass.
FinCorp's internal reporting tool runs on Linux in the non-production contract. It needs little CPU today, but its RAM need will grow over the next year as more reports are added, and the team wants to grow it without a migration. A team member proposes Cubes because they can attach Block Storage. What should the architect recommend?
The vCPU Server is the cheap class without CPU-family control, and on Linux it can grow RAM live, which matches a workload that grows in place. A Cube can attach up to 23 add-on Block Storage devices, but its vCPU, RAM and NVMe template cannot change after provisioning, so RAM growth would mean a new Cube, which is a migration. Confidential VM locks its RAM at creation, which is the opposite of the requirement.
FinCorp's portal servers in Frankfurt boot from a private golden image with cloud-init user data. The recovery VDC in Berlin must be able to provision identical servers. Which approach works?
Private images are usable only in the region where they were uploaded, and snapshots only in the location where they were created, so each location needs its own upload from a pipeline that stays the source of truth. Cloud-init can only be set at the initial provisioning of a volume, so it cannot be applied to restored, already provisioned volumes. Public images are rebuilt with each patch and the previous version leaves the catalog, so the same name does not guarantee the same image.
FinCorp's staging portal runs as a VM Auto Scaling group behind an ALB. The scale-in and scale-out thresholds are 25% and 70% CPU, the range is 120 seconds, and the cooldown is 2 minutes. New replicas need about 6 minutes to boot, run cloud-init and pass the ALB health check, and the group keeps adding and removing replicas. What is the best fix?
A cooldown shorter than the time a replica needs to become a healthy target lets the group measure again before the new capacity counts, which produces flapping; the range smooths short spikes before a decision. A group has exactly one metric policy, and no more than five VMs per action are recommended. VM Auto Scaling scales horizontally only.
The FinCorp team updated the replica configuration of the staging auto scaling group with a new image and more RAM. A week later, the running replicas still use the old image. How should the architect roll out the change?
Replica configuration changes apply only to new replicas, so the rollout happens through the scaling cycle, and an oldest-first termination policy removes replicas built from the old configuration first. Deleting the group deletes all its VMs, which turns a rollout into an outage. Deleting the newest replicas first would remove exactly the replicas that carry the new configuration.
FinCorp's core-banking systems run on VMware. The operations team must keep its vSphere tooling, the risk office requires single-tenant hardware in Germany, the migration must complete within one quarter, and storage is growing faster than compute. Which design should the architect recommend?
Private Cloud keeps the VMware operating model on dedicated hosts, VCDA migrates VMs with no charge for the migration feature, and an NFS datastore adds storage without adding hosts while the vSAN slack rule is respected. Backups, VM management and vSAN capacity are customer responsibilities, not part of the support scope, and a cluster needs at least three hosts. Rebuilding the estate on VDC compute gives up the VMware tooling, and VM Auto Scaling, which is in Early Access, scales replicas, not storage.