Skip to main content

Technical Lab: Manage Virtual Machine Sizes

Questions​

Question 1 — Multiple Choice​

A data team needs to run machine learning workloads with high GPU consumption. The environment is already in production and developers report that the current D-series VMs are experiencing parallel processing bottlenecks. Which Azure VM family is most suitable for replacement in this scenario?

A) E-series, optimized for memory
B) N-series, with dedicated GPU support
C) F-series, optimized for CPU computation
D) B-series, with accumulable CPU credits

Question 2 — Technical Scenario​

An administrator needs to resize a production VM from Standard_D2s_v3 to Standard_D8s_v3. When trying to execute the resize through the Azure portal, the confirmation button is disabled and the operation cannot be completed. What most likely explains this behavior?

A) The VM is using unmanaged disks, which don't support resizing
B) The target size is not available in the hardware cluster where the VM is allocated
C) The subscription has reached the vCPU quota for the region and blocked the operation
D) The VM is associated with an Availability Zone and cannot be resized without being recreated

Question 3 — True or False​

An Azure VM can be resized to any size available in the same region without the need for deallocation, as long as the new size belongs to the same family and generation as the current VM.

Question 4 — Technical Scenario​

A company runs a Standard_D4s_v3 VM and needs to migrate to Standard_E4s_v3 to meet new memory requirements. The administrator plans to execute the following flow:

1. Stop the VM (Stop)
2. Select the new size in the portal
3. Apply and start the VM

A colleague warns that simply clicking "Stop" in the portal is not sufficient to guarantee resizing between different families. What is the correct state that the VM needs to reach before resizing in this case?

A) Stopped (OS shut down, but still allocated on host)
B) Deallocated (released from physical host, with no hardware allocation)
C) Paused (suspended in memory, maintaining execution context)
D) Hibernated (state saved to disk with session preservation)

Question 5 — Multiple Choice​

When comparing the Standard_B2ms and Standard_D2s_v3 sizes for hosting an internal web application with unpredictable traffic and occasional spikes, which statement correctly describes the behavioral difference between the two sizes?

A) B2ms offers constant CPU performance, while D2s_v3 uses credits for spikes
B) D2s_v3 is ideal for loads with sporadic spikes, as it accumulates credits during idle periods
C) B2ms accumulates CPU credits when utilization is below baseline, allowing temporary performance spikes
D) Both sizes offer constant dedicated CPU, differing only in the amount of available memory

Answer Key and Explanations​

Answer Key — Question 1​

Answer: B

The N-series VMs (such as NC, ND and NV) are the only Azure families equipped with dedicated NVIDIA GPUs, designed for machine learning, deep learning and parallel computing workloads. Option B is the only technically defensible answer for the described scenario.

The distractors represent common confusions:

  • The E-series is optimized for memory, not for parallel processing with GPU
  • The F-series prioritizes the vCPU/memory ratio for CPU-intensive loads, without GPU
  • The B-series is for loads with low average CPU consumption and doesn't offer specialized hardware acceleration

Choosing the F or E series in an ML scenario would result in inadequate performance and high cost without the expected gain.

Answer Key — Question 2​

Answer: B

In Azure, VMs are allocated in physical hardware clusters. Not all sizes available in a region are present in all clusters. When the target size doesn't exist in the cluster where the VM is hosted, Azure blocks the operation because it cannot move the VM to another host without deallocating it.

The solution in this case is to deallocate the VM (Deallocate), which releases the binding with the current cluster and allows Azure to reallocate the VM to a host that supports the desired size.

The other distractors are plausible, but incorrect in this context:

  • Unmanaged disks can limit some features, but don't prevent resizing within the same family
  • vCPU quota generates a specific error message, doesn't silently disable the button
  • Availability Zones don't prevent resizing by themselves

Answer Key — Question 3​

Answer: False

The statement is false. The need for deallocation doesn't depend on family or generation, but on the availability of the target size in the physical cluster where the VM is allocated. Even within the same family, if the target size is not available on the current hardware, deallocation will be necessary.

Additionally, resizing between different families almost always requires deallocation because they use distinct hardware generations. The correct behavior is: if Azure indicates that the size is available without deallocation, the operation can be performed hot; otherwise, deallocation is mandatory regardless of family.

Answer Key — Question 4​

Answer: B

The Deallocated state is the only one that effectively releases the VM from the physical host in Azure. When a VM is only in the Stopped state (OS shutdown via portal or operating system), it remains allocated on the same hardware cluster, which prevents Azure from moving it to a different host that supports the new size.

This is especially relevant in migrations between families (such as D to E), which almost always require distinct hosts.

The other states described in the distractors don't exist as manageable states in the VM lifecycle in Azure as described: Paused and Hibernated are not standard management states for this resizing flow.

Answer Key — Question 5​

Answer: C

The B-series VMs work with a CPU credits model (burstable). When the VM operates below the CPU baseline defined for that size, it accumulates credits. These credits can be consumed during spikes, temporarily elevating CPU performance above the baseline.

This model is suitable exactly for loads with unpredictable traffic and occasional spikes, like the described web application.

The D-series offers dedicated CPU with constant performance, without credit accumulation, being more suitable for continuous and predictable loads.

Distractors A and B reverse the behavior of the two sizes, and distractor D incorrectly states that both offer constant performance, which completely ignores the B-series burstable model.