Skip to main content
Call Eric:863-698-8266
CURRYCONTROLS.COMControls & Automation Knowledge Hub
ReferenceSCADADesignCybersecurityNetworkingDocumentation

SCADA Servers

What the servers in a SCADA system do and what a utility should expect of them: the I/O, alarm, historian, and client roles, physical versus virtual, redundancy, sizing, placement, backup, patching, licensing, and monitoring.

10 min readUpdated Sep 5, 2026Published Sep 5, 2026By Eric Sullivan

The short answer

SCADA Servers

A SCADA system runs on one or more servers that poll the controllers, evaluate alarms, store history, and serve displays to clients, with the roles combined on one machine in a small system and separated on several in a larger one. Servers are increasingly virtual machines on a hypervisor, which simplifies backup and hardware replacement, and critical systems run redundant pairs with automatic failover. They are sized by tag count, poll rate, client count, and history retention, placed in the supervisory zone with any web or remote access function in a DMZ, kept in a cooled room on a UPS, backed up as both an image and an application export, patched on a schedule, and monitored for disk, memory, service health, and time synchronization.

Key points

  • Four roles: talk to the controllers, evaluate alarms, keep history, serve displays. Combine or split them by size.
  • Virtual machines are the norm now; the hypervisor host and its storage are the hardware to worry about.
  • Redundancy is a pair with automatic failover and a client that follows; test the failover, not just the hardware.
  • Size for tags, poll rate, clients, and retention, with headroom for the next expansion.
  • Supervisory zone for the servers, DMZ for anything a remote user touches, a cooled room and a UPS for all of it.
  • Back up the image and the application export; patch on a schedule; monitor disk, memory, services, and time.

The roles

RoleWhat it doesNotes
I/O or tag serverPolls controllers and devices through drivers or OPC, holds the current value and quality of every tag, and serves them to everything elsePoll load and driver licenses live here; the first thing to size
Alarm serverEvaluates alarm conditions, manages alarm states and acknowledgments, journals alarm events, drives notificationOften the same process as the tag server on small systems
HistorianStores time-series data with compression and serves trends and reportsDisk and retention are the sizing items; can be a separate product
Client or web serverServes displays to thin and web clients, runs scripts and reportsClient count and script load; the one a remote user connects to, so it belongs in a DMZ when exposed
DatabaseRelational storage for alarm journals, audit trails, configuration, and reportsBacked up with the application; a database that fills the disk stops the system
Supporting servicesTime synchronization, directory and authentication, remote access gateway, backupUtilities that the SCADA depends on but that are not SCADA

Physical or virtual

A physical server per role was the design for years and still exists in small systems and in vendor packages. Most new systems run each server as a virtual machine on one or two hypervisor hosts. The advantages are real: a virtual machine is backed up as a file, restored to different hardware in minutes, snapshotted before a patch, and moved between hosts for maintenance. The costs are that the hosts and their shared storage become the critical hardware, that the hypervisor needs its own patching and skills, and that a single host running everything is a single point of failure dressed up as several servers. Two hosts with the redundant pair split between them, on separate storage or with replication, is the arrangement that keeps the redundancy real.

Redundancy

A redundant pair runs the same configuration on two servers; one is active, the other is standby and synchronized, and the clients switch to the standby when the active fails. The details differ by product: some pairs both poll the controllers, some have the standby poll only after failover, some synchronize alarm acknowledgments and some do not. Whatever the product does, the test is to pull the power on the active server during normal operation and watch: clients reconnect within the time the philosophy allows, alarms continue with their states intact, the historian has no gap or a gap that is backfilled, and when the failed server returns it resynchronizes without taking over. Redundancy that was configured and never tested is a second server.

Sizing

  • Tags: the number of points polled, which sets memory, driver load, and license tier. Count the tags in the controllers, not the ones on the screens.
  • Poll rate: how often each device is read. A thousand tags at one second is a different load from ten thousand at ten seconds, and radio-linked sites poll slowly by necessity.
  • Clients: each connected client adds display rendering and script load on the client server.
  • History: tags historized, rate, compression, and retention decide disk. Estimate bytes per sample times samples per day times days, then double it.
  • Headroom: the next plant expansion, the next ten lift stations, and the tag growth that every upgrade brings. Fifty percent spare is not extravagant.

Placement

The servers belong in the supervisory zone of the network, behind the firewall that separates them from the controllers, with the controllers reachable only through the rules that the polling needs. Anything a user outside the control room touches, a web server, a remote access gateway, a reporting portal, sits in a DMZ so that a compromised remote session reaches the DMZ and not the servers. Physically, they belong in a room with cooling, a UPS sized for the generator start time, clean power, restricted access, and no water pipes overhead, which rules out most plant floor panels and many electrical rooms. A server in a panel next to a drive lives a short and unreliable life.

Care and feeding

Backup
An image of the virtual machine or the disk, on a schedule, kept off the server and off site, plus an export of the application configuration in the vendor format, so that the project can be restored on a new version of the software. Test a restore once a year.
Patching
Operating system and application patches on a schedule, after the vendor has qualified them, applied to the standby first, with a snapshot before each and a rollback plan.
Licensing
Tag counts, client counts, driver and historian licenses, and the keys or dongles that unlock them. Keep the license files and activation records with the backup; a restored server with no license is a restored server that will not start.
Monitoring
Disk space, memory, CPU, service status, poll success rate, historian queue depth, backup completion, and time synchronization, alarmed to the same place as process alarms. A full disk is the most common cause of a dead SCADA server, and it announces itself for weeks.
Lifecycle
Hardware replaced on a five to seven year cycle, operating systems before they leave support, and the application on the vendor supported versions. A server that cannot be patched is a security finding and an availability risk.

Frequently asked questions

Can a small utility run everything on one server?
Yes, and many do: one machine polling a handful of sites, evaluating alarms, keeping history, and serving two clients. Make it a virtual machine so that it can be backed up and restored easily, put it on a UPS in a decent room, and keep a tested restore. Redundancy comes when the cost of an outage justifies it.
Should the historian be on the same server as the SCADA?
On small systems, yes. On larger ones the historian is separated because its disk and query load interferes with polling and alarms, and because the historian is what reporting and outside users access, which is better done away from the server that controls the plant.
How long should the UPS run?
Long enough for the generator to start and stabilize, with margin, and long enough for a clean shutdown if the generator does not. Fifteen to thirty minutes is common. The UPS also feeds the network switches, the firewall, and the time server; a server with power and no network is not much use.
What does time synchronization have to do with servers?
Every alarm, event, and history sample carries a timestamp, and clients, servers, controllers, and the historian have to agree on the time or the records cannot be compared. One time source, usually a GPS-disciplined or network time server, synchronizes all of them, and a server that drifts is alarmed.

Direct contact

Have a controls question?

Reach Eric Sullivan directly about anything on this site, a controls or automation topic, or one of his personal projects.