OpenStack — cloud toolkit that grew into an ecosystem
What it is
OpenStack started as a joint effort (NASA + Rackspace) and over the years turned into a whole collection of projects bundled together. It isn’t a single product — more like a toolbox for building a private or public IaaS cloud. At its heart it spins up VMs, wires up virtual networks, and attaches storage, but once you start digging you find dashboards, APIs, orchestration engines, image catalogs, identity services… the list goes on. That flexibility is why big providers and telcos still run it, even if the hype days are over.
How it works (in practice)
– Nova schedules and launches VMs, usually on KVM.
– Neutron handles networks — VLANs, VXLANs, GRE tunnels, firewalls, plugins for SDN.
– Cinder attaches block volumes, while Swift gives you S3-like object storage.
– Glance is the image catalog, Keystone the central identity service.
– Horizon is the dashboard, though most real work happens via REST APIs.
– Heat orchestrates whole stacks — multiple services, networks, storage in one template.
Technical map
| Area | Notes |
| Type | IaaS platform (modular) |
| Hypervisors | KVM by default, also Xen, Hyper-V, VMware |
| Networking | Neutron (VLAN, VXLAN, GRE, SDN plugins) |
| Storage | Cinder (block), Swift (object), Manila (file), Ceph integration common |
| Identity | Keystone |
| Dashboard | Horizon |
| APIs | REST across all services |
| License | Apache 2.0 |
| Scale | One-node labs up to telco-scale clouds |
Deployment notes (from real setups)
– A “full” cloud = multiple controllers, compute nodes, and separate storage.
– For labs there’s DevStack, which crams everything on one host — useful for demos, not production.
– Ceph is the usual backend for block and object storage.
– Network design is often the hardest part: provider vs tenant networks, overlays, routers, NAT.
– Almost every component is API-driven; Horizon is convenient but not mandatory.
– Upgrades can be painful — rolling through versions takes planning.
Where it fits
– Telcos building NFV/edge clouds.
– Service providers offering multi-tenant IaaS.
– Research centers handing out self-service VMs to teams.
– Large enterprises that want VMware-like functionality but under their control.
Weak spots
– Heavy: not something you casually drop on three servers and call it done.
– Needs staff who understand both networking and distributed systems.
– Smaller companies often find it overkill compared to Proxmox or vSphere.
– Community is still alive, but not as loud as it was ten years ago.
Comparison snapshot
| Tool | What makes it different | Best suited for |
| OpenStack | Full IaaS stack, modular | Telcos, big enterprises |
| vSphere (ESXi + vCenter) | Polished, vendor-backed | Enterprises, corporate IT |
| Proxmox VE | Simple, community-driven | SMBs, labs |
| oVirt | RH-aligned, centralized control | RHEL/CentOS shops |
Quick start sketch
1. Install Linux (Ubuntu, Rocky, CentOS).
2. Use DevStack for lab deployment.
3. Log into Horizon, spin up a small VM.
4. Play with Neutron networks, attach a Cinder volume.
5. For production, split roles across controllers, compute, and storage nodes.
Field notes — 2025
– Still entrenched in telco space; many 5G rollouts use it underneath.
– Pairing with Ceph is almost the default choice.
– Complex to operate — requires people who “live” in it, not occasional admins.
– Kubernetes can be deployed inside or alongside, but many treat them as separate layers.
– For small shops, Proxmox or VMware is simpler; OpenStack only shines when you scale big.