Running a SaaS on bare metal

There is a persistent myth that a serious SaaS must run in a public cloud. In fact a large number of healthy software businesses run on bare metal, and discover their infrastructure bill is both lower and — crucially — predictable. The trick is a sane architecture, not a magic platform.

A shape that works

• A reverse proxy at the front for TLS and routing
• Stateless application servers you can add more of — see spreading traffic across two servers
• A solid database with a replica — see replication
• Object storage for user files, Redis for cache and sessions

Predictable beats elastic

The cloud sells elasticity; most SaaS workloads are far more predictable than that pitch assumes. If your load has a known daily shape, a fixed set of dedicated servers sized for the peak is cheaper than paying a premium for elasticity you rarely use — and the bill does not surprise you. See dedicated, VPS or cloud?.

Keep state in known places

The thing that makes scaling hard is state scattered everywhere. Keep it deliberately in the database, the cache and object storage; keep the application servers stateless. Then growth is just adding another identical app server behind the proxy — no drama.

Scale by addition

When one app server is not enough, add a second and let the proxy share traffic. When the database is the limit, add read replicas and a pooler. Most SaaS growth is handled this way long before anything exotic is needed — see capacity planning.

Resilience to match the stakes

A SaaS has customers who notice downtime, so build to your real cost of downtime: tested backups, a replica, and a second app server remove most risk cheaply. Tell us where you are heading and we will size the path — see capacity planning.

---

Still stuck? Mail support@novogara.com — an engineer answers, at any hour. Back to the knowledge base