Compute

Compute Engine

Virtual machines running in Google's data centers

AWS equivalent

EC2

IaaSVMsCore Service

🔄

AWS → GCP: Key Differences

▸
Machine types use a different naming convention: e2 (cost-optimized), n2 (balanced), c2 (compute), m1 (memory). No 'instance families' like t3/m5.
▸
Preemptible VMs ≈ Spot Instances — up to 80% cheaper, can be terminated with 30s notice. Spot VMs are newer with similar but slightly different behavior.
▸
Sustained use discounts apply AUTOMATICALLY after running 25%+ of a month — no reserved instance purchase needed.
▸
Live migration: GCP migrates your VM to another host during maintenance without rebooting. AWS reboots.
▸
Custom machine types: create exactly the vCPU/RAM combination you need instead of picking from fixed sizes.

📌

1
Managed Instance Groups (MIGs) = Auto Scaling Groups in AWS. Can be regional (multi-zone) or zonal.
2
Startup scripts = EC2 user data. Can be stored in GCS and referenced at launch.
3
Persistent disks (pd-standard, pd-balanced, pd-ssd) ≈ EBS. Local SSDs are ephemeral but extremely fast.
4
Sole-tenant nodes: run VMs on dedicated physical hardware — for compliance or licensing reasons.
5
OS images: public (Debian, Ubuntu, CentOS, Windows) or custom. Can be shared across projects.

💡

★
Know when to recommend Compute Engine vs Cloud Run vs GKE — use Compute Engine for workloads needing full OS control, legacy apps, or specific software that can't be containerized.
★
Understand MIG autoscaling policies: CPU utilization, HTTP load balancing, custom metrics.
★
Be able to explain cost optimization: preemptible VMs for batch, committed use discounts for stable workloads, rightsize recommendations.

⚠️

!
GCP charges per-second (after 1 min minimum) vs AWS per-hour historically — better for short-running workloads.
!
External IPs cost money even when not in use if reserved. Release them when not needed.
!
Default service account has broad Editor role — always create dedicated service accounts with least privilege.