Compute Engine

Overview

• Highly configurable
• Zonal
• Choice of machine types:
  • General purpose
  • Compute optimized
  • Memory optimized
  • Processor optimized (GPUs)
• Public or private boot disk images
• Live migrations
• Auto restart
• Per second billing—minimum of one minute
  • vCPU and memory billed separately

Boot Disks

• Set --no-boot-disk-auto-delete to keep boot disk data after instance deletion

Preemptible VMs

• Can stop at any time
• Will definitely stop after 24 hrs
• 60 – 90% cheaper than standard VMs

Sole Tenant Nodes

• Dedicated server
• Useful for specific hardware requirements, e.g. for bring your own licence (BYOL) requirements—per processor/core licencing

Managed Instance Groups (MIGs)

• Multiple identical VMs managed together
• Autoscaling
• Auto-healing
• Regional/multi-zonal
• Unmanaged Instance Groups:
  • Load balance against any group of VMs

Machine Families

Use Cases

Family	Use case
General purpose (E2/N2/N2D/N1)	Balanced performance and cost
Compute optimized (C2/C2D)	High-performance, CPU frequency and consistency, powerful cores, high CPU/mem ratios
Memory optimized (M1/M2)	High memory requirements
Accelerator optimized (A2)	Highest performance GPUs for ML, HPC and parallel computing
Scale-out optimized (Tau T2D)	Horizontally scalable cloud native workloads

GPUs

• GPUs supported on:
  • N1
  • A2

Images

Custom Images

• Created from: source disks, images, snapshots, images stored in Cloud Storage
• Used to create VM instances
• Virtual disk import tool: import boot disk images from existing systems
• Stored in Cloud Storage—zonal or regional
• Can group into families
• Image status:
  • ACTIVE—image can be used as normal
  • DEPRECATED—can still be used, but marked as deprecated
  • OBSOLETE—can no longer be used
  • DELETED—can no longer be used

References

• Google Cloud Decision Trees