GCP & Azure Fundamentals for Multi-Cloud

GCP Core Services: Compute, Storage & Networking

4 min read

Google Cloud Platform excels in data analytics, Kubernetes, and machine learning. Understanding GCP is essential for multi-cloud architect roles.

GCP Compute Services

Compute Engine (VMs)

GCP's equivalent to AWS EC2 with some unique features.

Machine Type Families:

Family Use Case Example
General Purpose (E2, N2) Web serving, dev/test e2-standard-4, n2-standard-8
Compute Optimized (C2, C3) HPC, gaming, batch c2-standard-30
Memory Optimized (M2, M3) SAP HANA, in-memory DBs m2-ultramem-416
Accelerator Optimized (A2, A3) ML training, HPC a2-highgpu-8g

Unique GCP Features:

  • Preemptible VMs: Up to 91% discount (vs. 90% Spot on AWS)
  • Sustained Use Discounts: Automatic discounts for running instances (no commitment)
  • Custom Machine Types: Configure exact vCPU and memory

Google Kubernetes Engine (GKE)

GKE is often considered the best managed Kubernetes service.

Why GKE Stands Out:

  • Autopilot Mode: Fully managed, pay per pod
  • Release Channels: Rapid, Regular, Stable
  • Multi-cluster Ingress: Global load balancing across clusters
  • Built-in Istio: Service mesh integration

Interview Question: GKE vs EKS

Q: "Compare GKE Autopilot to EKS Fargate."

A:

Feature GKE Autopilot EKS Fargate
Management Fully managed (nodes, scaling) Serverless pods only
Pricing Per pod (vCPU + memory) Per pod (vCPU + memory)
Node Access No (abstracted) No (abstracted)
GPUs Yes (with limitations) No
DaemonSets Supported Not supported
Best For Teams wanting minimal ops Serverless-first workloads

Cloud Functions & Cloud Run

GCP's serverless compute options.

Cloud Functions:

  • Event-driven, similar to AWS Lambda
  • Max timeout: 60 minutes (2nd gen) vs. 15 min Lambda
  • Support for HTTP and event triggers

Cloud Run:

  • Container-based serverless (unique to GCP)
  • No cold starts for always-allocated instances
  • Full HTTP/gRPC support
  • Request-based pricing or always-on

When to Use Each:

Simple event processing → Cloud Functions
Containerized workloads → Cloud Run
Long-running containers → Cloud Run (always-allocated)
REST/gRPC APIs → Cloud Run

GCP Storage Services

Cloud Storage (Object Storage)

Storage Classes:

Class Minimum Duration Retrieval Use Case
Standard None Immediate Frequently accessed
Nearline 30 days Immediate Monthly access
Coldline 90 days Immediate Quarterly access
Archive 365 days Immediate Yearly access

Key Difference from S3: All GCP classes have immediate retrieval (unlike Glacier).

Persistent Disk Types

Type Max IOPS Max Throughput Use Case
pd-standard (HDD) 7,500 400 MB/s Bulk storage
pd-balanced (SSD) 80,000 1,200 MB/s General workloads
pd-ssd (SSD) 100,000 1,200 MB/s High performance
pd-extreme (SSD) 120,000 2,200 MB/s Databases, SAP

Filestore (Managed NFS)

GCP's equivalent to AWS EFS.

Tiers:

  • Basic: General file sharing
  • Enterprise: High availability, snapshots
  • High Scale: Big data workloads (100+ TB)

GCP Networking

VPC Concepts

GCP VPC Differences from AWS:

  • Global VPCs: VPCs span all regions (not regional like AWS)
  • Subnets are regional: Span all zones in a region
  • Firewall rules: Applied at VPC level, not subnet level
  • Shared VPC: Cross-project networking

Interview Question: GCP vs AWS VPC

Q: "What are the key differences between GCP and AWS VPC design?"

A:

Aspect GCP AWS
VPC Scope Global Regional
Subnet Scope Regional (all zones) AZ-specific
Firewall VPC-level rules Security Groups + NACLs
NAT Cloud NAT (regional) NAT Gateway (per AZ)
Peering Global, transitive routes Regional, non-transitive
Private Access Private Google Access VPC Endpoints

Cloud Load Balancing

GCP's global load balancing is a key differentiator.

Load Balancer Types:

Type Scope Protocol Use Case
HTTP(S) LB Global Layer 7 Web apps
TCP Proxy Global Layer 4 TCP apps
SSL Proxy Global Layer 4 SSL termination
Network LB Regional Layer 4 Low latency
Internal HTTP(S) Regional Layer 7 Internal services
Internal TCP/UDP Regional Layer 4 Internal apps

Why Global Load Balancing Matters:

  • Single anycast IP serves all regions
  • Automatic routing to nearest healthy backend
  • No need for Route 53-style DNS failover

Interview Tip: GCP's global network is its biggest differentiator. Emphasize premium tier networking and global load balancing when comparing to AWS.

Next, we'll explore GCP's data and AI services. :::

Quiz

Module 3: GCP & Azure Fundamentals for Multi-Cloud

Take Quiz