Regions and availability zones
Cloud infrastructure is physically organized in a hierarchy:
- Region — a geographic area where a provider clusters data centers (e.g. "US East")
- Availability zone (AZ) — one or more isolated data centers within a region, each with its own power, cooling, and networking
- Data center — racks, servers, storage, and networking equipment
Spreading workloads across AZs contains failures (one AZ going down doesn't take others with it) and lets you place resources physically closer to users to cut latency.
Compute options
| Option | Description | Best for |
|---|---|---|
| Virtual machine | A hypervisor slices one physical server into many isolated VMs | General-purpose workloads |
| Bare metal | A single-tenant, dedicated physical server | Compliance, licensing, max performance |
| Container | Packages code + dependencies, shares the host OS kernel | Fast, portable, lightweight deployment |
| Serverless | You upload functions; the platform runs, scales, and bills them | Event-driven, spiky workloads |
A hypervisor is what makes VMs possible: Type 1 ("bare-metal") runs directly on the hardware for lower latency and better security (VMware ESXi, Hyper-V, KVM); Type 2 ("hosted") runs on top of a host OS, trading some performance for convenience (VirtualBox, VMware Workstation).
Networking in the cloud
Cloud networks are built from the same primitives as on-premises networks, just software-defined:
- Public interfaces connect resources to the internet; private interfaces connect them to each other, kept off the public internet
- IP addresses and subnets carve up the address space assigned to your network
- Security groups and ACLs control exactly which traffic is allowed in or out
- VPCs and VPNs isolate your resources and extend your private network into the cloud
- Load balancers and gateways distribute traffic and manage entry points
Getting this right is mostly a question of "who and what can reach this resource" — the fewer things exposed to the public internet, the smaller the attack surface.
Four types of cloud storage
| Type | How it works | Good for |
|---|---|---|
| Direct-attached | Local disk on the same server/rack | OS disks, temporary/ephemeral data |
| File storage (NFS) | A shared network filesystem, mountable by many nodes | Shared file access, departmental shares |
| Block storage | Data split into addressable blocks, mounted as a volume | Databases, mail servers — anything needing low, consistent latency |
| Object storage | Flat storage accessed via API, organized into buckets | Static files, backups, archives, VM images |
Direct-attached storage is fastest but ephemeral and non-shareable; object storage is the opposite — nearly infinite and cheap, but the slowest to access.
Choosing the right storage tier
A few metrics decide which storage type and tier fits a workload:
- IOPS (input/output operations per second) — higher IOPS means faster reads/writes, but costs more; too low and it becomes a bottleneck
- Snapshots — fast, point-in-time backups that only record what changed, though they can't restore a single file in isolation
- Storage tiers — standard (frequent access, highest cost per GB), archive/vault (rarely accessed, lowest cost, slower retrieval)
- CDNs — cache content geographically closer to users, cutting load times and reducing origin traffic
Tip: picking a storage tier is really a bet on how often the data will be read — bet wrong, and you either overpay for speed you don't need or wait hours to retrieve something urgent.