How-to
How to compare SCADA historian functions: a test method
Short answer
There is no maintained public function-level comparison of SCADA historians, so build your own. Define matrix rows for compression, store-and-forward, alarm and event history, external access, retention and extensions; run identical tests on each candidate's recorded version; score pass, fail or not tested. Eliminate any candidate failing a mandatory row and rank survivors on engineering effort.
On this page
- Is there a maintained public comparison of SCADA historian functions?
- Which historian functions belong in the matrix?
- How do I write a requirement that can pass or fail?
- How do I run the same test on every candidate?
- How much disk does a 1-second historian need for 400 tags?
- How do I score the results and choose?
- When is the SCADA's own logging no longer enough?
- Questions asked
Build a matrix of 10 to 20 rows covering seven historian functions — compression, filtering, store-and-forward buffering, separate alarm and event history, external retrieval, retention management and the extension mechanism — then run one identical test procedure on every candidate's recorded version and score each cell pass, fail or not tested, with any mandatory-row fail eliminating the candidate outright. A worked sizing check for 400 tags at 1-second logging on a raw 14-byte record gives 0.484 GB/day, 177 GB over a full year and roughly 900 GB for five years of raw retention, so plan a 1 TB volume.
There is no maintained, vendor-neutral comparison of SCADA historian functions anywhere on the public web, and the matrices that circulate are vendor marketing that goes stale within one release cycle, so the matrix you build and test yourself is the only defensible source. This article gives that method end to end: the historian functions that belong in the matrix, how to phrase a requirement so it can fail, the test procedure, the disk-sizing arithmetic, the scoring rule, and the line where a standalone historian beats the SCADA's own logging.
Is there a maintained public comparison of SCADA historian functions? #
No. No vendor-neutral body maintains a function-level comparison of SCADA historians that survives even one release cycle, so the only defensible source is a matrix you build and test yourself. What you find online instead are point-in-time tables that name products without versions, and vendor matrices that count checkboxes rather than outcomes; a comparison that cannot say which version it tested is not evidence about the version you run.
A shortlist can mix current lines such as InTouch or WinCC with retired packages such as FactoryLink or Paragon; the method below does not change, but a package no longer sold cannot pass a support row, so record that fact in the matrix rather than leaving the row blank. If the decision sits one level up — which platform at all — work through how to choose HMI software for a new control system first; everything below assumes the shortlist exists and asks which historian functions each candidate actually delivers.
Which historian functions belong in the matrix? #
Put seven functions in the matrix: compression, filtering, store-and-forward buffering, separate alarm and event history, external retrieval, retention management and the extension mechanism, because those are the rows that decide a historian comparison.
| Historian function | What the row asks | Evidence you record |
|---|---|---|
| Deadband compression | Does the archive skip values that stay inside a set band? | Stored record count for the same replay with and without the band |
| Filtering | Can a query return interpolated values at fixed intervals? | Interpolated output at 1-minute intervals beside the raw samples |
| Store-and-forward | Does logging continue at the source when the path to the archive drops? | Sample count before, during and after a timed network cut |
| Alarm and event history | Are alarm states and operator actions stored separately from process values? | Exported entries with timestamps and user attribution |
| External access | Can a third-party client read the archive without vendor middleware? | Query result through OPC UA historical access or direct SQL |
| Retention management | Does data purge or archive automatically at a set age? | The configuration that sets the age boundary, and the file count after it |
| Extension | Can project-specific calculations run against logged data? | A working totalizer or rollup running on the logged tags |
Compression and filtering decide most comparisons: a product that cannot skip values inside a deadband and then answer queries with interpolated values will not keep years of history at a usable size, and that pair is what qualifies software as a process historian rather than a logger. External access should be concrete: OPC UA includes historical access in the specification, so a compliant third-party client reads the archive without vendor middleware, and the OPC UA companion model for ISA-95 (specification OPC-10030) defines the production objects higher-level systems expect. If operators must see history while the network to a central archive is down, add one more row for local logging at the panel, because products differ exactly there.
How do I write a requirement that can pass or fail? #
Write every row as an operation with a measurable result and a fixed input, and cap the matrix at 20 rows with no more than a third marked mandatory, or the test campaign never finishes. A row is testable when it names the operation, the input and the measured result. 'Good historical trending' is a brochure line; 'a query over any 8-hour window of the tag set returns in under 5 seconds on the test hardware' is a requirement, and the 5 seconds is your acceptance line — set it where operators would start complaining, then hold every candidate to the identical number. Mark each row mandatory or preference in writing before the first test, so a late favourite cannot demote a failure to a nice-to-have. Treat the mandatory set as elimination rules and keep it small, because a matrix where everything is mandatory eliminates candidates on trivia.
How do I run the same test on every candidate? #
Run one identical procedure against every candidate on a recorded version, with the same tag list, the same scan rate and the same scripted queries, so any difference in results is a product difference.
- Record the candidate's identity before anything else: product, exact version, installed modules, operating system and patch level, filed with the results.
- Load the same tag list and the same scan rates on every candidate, driven by a replayed data set or a simulator.
- Configure the historian the same way each time: identical deadband, identical logging interval, identical store-and-forward settings.
- Run the scripted retrieval tests — a fixed time-range query, a fixed-interval aggregate and one export — and time each.
- Cut the network path to the archive for 10 minutes while values keep changing, restore it, then count the samples present against what the source produced.
- Capture evidence as you go: configuration screens, exact query text, result counts and timings.
- Mark each row on the day you run it, and record why any row stayed open.
- Re-run a row that failed for environmental reasons — a full disk, a wrong address entry — before it counts as a product failure.
Run every step on a development system or an offline copy: on a live plant, a cut network path or an unbounded query is a production event, not a test.
How much disk does a 1-second historian need for 400 tags? #
At 400 analog tags logged once per second, a raw record of 14 bytes grows the archive by about 0.48 GB over a day and 177 GB over a full year, so five years of raw retention needs roughly 900 GB. Illustrative sizing on a raw, uncompressed record of 14 bytes — an 8-byte timestamp, a 4-byte IEEE 754 floating-point value and 2 bytes of quality and status:
- Samples: 400 tags × 1 sample/s = 400 samples/s
- Write rate: 400 samples/s × 14 B/sample = 5,600 B/s
- Over one day: 5,600 B/s × 86,400 s = 483,840,000 B = 0.484 GB, taking 1 GB as 1,000,000,000 B
- Over 365 days: 0.484 GB × 365 = 176.7 GB, call it 177 GB
- Five years of raw retention: 176.7 GB × 5 = 883 GB
Plan a 1 TB volume: it covers the raw payload with headroom for indexes, logs and tag-count growth. A 500 GB volume holds 500 ÷ 176.7 ≈ 2.8 years raw, so accept the shorter window, take the larger volume, or let deadband compression cut the write rate — then measure the reduction on your own data rather than trusting a vendor's ratio. This is payload arithmetic on an illustrative record size, not any product's footprint; after 24 hours of real logging, compare measured file growth against the projection. That measured check is the same discipline described in verifying HMI historical trend retention before committing.
How do I score the results and choose? #
Score every cell pass, fail or not tested, eliminate any candidate that fails a mandatory row outright, and rank the survivors on the engineering effort each row took, never on a marketing feature list. Define the states tightly. Pass means you observed the result yourself; fail means you observed its absence on a correctly configured system; not tested means the row never ran, and it is neither a zero nor a pass. Do not close the comparison with an open mandatory row — an unrun test defers the decision, it does not average out.
Keep capability and effort in separate columns: 'needed a script and two hours' is a different fact from 'cannot be done', and merging them is how a workable product loses to a polished demonstration. Rate effort per row on your own 1-to-3 scale and total it. The rule at the end is fixed: any fail on a mandatory row eliminates the candidate outright; among survivors, the lowest effort total wins; break a tie on the retrieval path your reporting stack actually uses.
When is the SCADA's own logging no longer enough? #
Keep logging in the SCADA when retention is under a year, tag counts stay in the hundreds and nobody audits the trail, and move to a standalone process historian the moment multi-year retention, audit trail or thousands of points become mandatory. The logging built into an HMI or SCADA package answers operator questions about recent production; it was not built for compliance-grade retention or audit trail, and standalone historians carry exactly those obligations while holding thousands of points where a panel-level logger holds hundreds. If operators must still see history while the network to the central archive is down, keep the local-logging row in play for whatever remains at the panel.
Give any standalone candidate the same rows and the same link-cut test, because a historian that silently loses data across a network drop is worse than the modest logging you already trust. Commit at the lines: under a year of retention, no audit obligation and tag counts in the hundreds, stay with the SCADA's own logging; cross any one of those lines and run the matrix for a standalone historian.
Questions asked
Can a vendor's published feature list count as evidence in the matrix?
No. Treat it as a claim to verify, not a result. Ask the vendor for a time-limited evaluation licence of the full product, run your mandatory rows yourself on your own tag list, and file the vendor document beside your measured outcome. Where the two disagree, your measured result sets the row's score.
How many rows should the comparison matrix have?
Between 10 and 20 rows, with mandatory rows capped at about a third of the total. Below ten you are only testing marketing bullet points; above twenty the test campaign outlives the evaluation window and versions change under you. If a stakeholder adds rows late, trade one out rather than growing the matrix past twenty.
Does a historian replace the plant's relational database?
No. A historian is built for one thing: thousands of time-stamped series written fast and queried by time range. Batch records, recipes, user accounts and configuration-change audit belong in a relational database, and most plants run both side by side. Put a row in the matrix only when a product claims to cover both, and test that claim.
Where do OPC UA and ISA-95 fit in the comparison?
As access and vocabulary rows, not as products. OPC UA historical access is the standard route by which an outside client reads an archive without vendor middleware, so it makes a clean pass-or-fail row. ISA-95, carried into OPC UA by the companion model in specification OPC-10030, defines the production objects higher-level systems expect, which matters when the historian feeds reporting above the control layer.