Core Google Cloud services

A short tour

Google Cloud has well over a hundred services, but a small set covers most applications. This lesson explains what each core service does, when to use it, and the one or two billing facts that matter. If you know AWS or Azure already, the next lesson maps these services to their equivalents.

Compute Engine: virtual machines

Compute Engine gives you virtual machines (VMs) in any region. You pick a machine type, which sets the number of vCPUs (virtual processors) and the memory. Machine types come in families: general-purpose (for example the low-cost E2 series, good for web servers and small databases), compute-optimised (for heavy calculation), memory-optimised (for large in-memory databases) and accelerator-optimised (with GPUs for machine learning). An e2-medium, for example, has 2 shared vCPUs and 4 GB of memory.

Spot VMs are much cheaper, but Google can stop them at any time when it needs the capacity. They suit batch jobs that can restart. A managed instance group (MIG) runs many identical VMs from a template and can add or remove VMs automatically as load changes.

Cloud Storage: objects in buckets

Cloud Storage is object storage: you store files (objects) in buckets. Bucket names are globally unique. A bucket has a location (one region, a dual-region or a multi-region) and a default storage class. All classes have the same fast access and the same API. The difference is price: colder classes cost less per GB stored, but charge for reading data back and have a minimum storage duration. Delete an object early and you still pay for the minimum.

gcloud storage buckets create gs://acme-invoices-2026 --location=asia-south1 --default-storage-class=STANDARD
gcloud storage cp invoice-1042.pdf gs://acme-invoices-2026/2026/10/
gcloud storage ls gs://acme-invoices-2026/2026/10/

Networking: a global VPC and Cloud Load Balancing

A VPC (Virtual Private Cloud) is your private network in the cloud. In Google Cloud a VPC network is global, and its subnets are regional. One VPC can have a subnet in Mumbai and another in Delhi, and VMs in both talk over private IP addresses without any extra peering. This is different from AWS, where a VPC belongs to one region. Firewall rules are set on the VPC network.

Cloud Load Balancing spreads traffic across your backends. The global external Application Load Balancer gives you one public IP address that serves users worldwide and sends each request to the nearest healthy backend. Cloud CDN caches content close to users, and Cloud DNS hosts your domain names.

Serverless: Cloud Run, Cloud Run functions and App Engine

Cloud Run runs any container that listens for HTTP requests. You do not manage servers. It starts more copies (instances) when traffic grows and scales down to zero when nobody is calling, so a quiet service can cost almost nothing. Cloud Run jobs run containers that do a task and finish, such as a nightly report.

Cloud Run functions (called Cloud Functions until 2024) let you write just a function, for example in Python or Node.js. Google builds and runs the container for you. They suit small event handlers, such as resizing an image when it lands in a bucket. App Engine is Google's older platform for web apps. It is still supported, but many new projects choose Cloud Run.

gcloud run deploy hello --source . --region asia-south1 --allow-unauthenticated

Databases

• Cloud SQL: managed MySQL, PostgreSQL and SQL Server. Google handles backups, patches and failover. The default choice for a normal web app.
• Firestore: a serverless NoSQL document database. It stores JSON-like documents and syncs well with mobile and web apps.
• Spanner: a relational database that scales horizontally across regions with strong consistency. Powerful, and priced for large systems.

BigQuery: the serverless data warehouse

A data warehouse is a database built for analysis: big scans, totals and trends over millions or billions of rows. BigQuery is serverless, so you write SQL and Google finds the machines. Storage and compute are billed separately. For compute you choose on-demand pricing, where you pay per TiB (tebibyte, 1024 to the power 4 bytes) of data your queries scan, with a monthly free amount, or capacity pricing, where you reserve processing power (slots) for a steady price.

BigQuery stores data by column. A query that selects 3 columns reads only those 3 columns, so SELECT * on a wide table costs far more than naming the columns you need. Adding LIMIT 10 does not reduce the data scanned on a normal table. Partitioning a table by date lets a query that filters on date read only the matching days.

bq query --use_legacy_sql=false --dry_run 'SELECT city, SUM(amount) FROM shop.orders WHERE order_date = "2026-10-01" GROUP BY city'

Pub/Sub and GKE

Pub/Sub is a messaging service. Publishers send messages to a topic, and each subscription delivers them to a subscriber. It decouples parts of a system: the order service publishes "order placed" and the email, invoice and analytics services each react in their own time. GKE (Google Kubernetes Engine) runs Kubernetes for you in Standard or Autopilot mode; module 8 covers it.

Free tier

Google Cloud has a Free Program with two parts: trial credit for new customers to spend in the first months, and an always-free tier with monthly limits on some products (for example a small e2-micro VM in certain US regions and a monthly amount of BigQuery queries). The amounts and rules change, so check the current Google Cloud Free Program page before relying on them.