Specifications¶
A specification is your current estate as Interact DC understands it: one line per server configuration, with the counts, CPUs, memory and any other detail you supplied.
Each line is a specification line, and represents one or more identical servers — fifty identical R750s are one line with a quantity of fifty, not fifty lines.
The three pieces of a simulation¶
It helps to keep these apart, because the app uses the words precisely.
Specification — what you run today. Built by CSV upload, by hand, or by importing a location.
Targets — the hardware you are considering instead. See Targets and runs.
Runs — one execution of the analysis, comparing the specification against every target. A simulation keeps all of its runs.
Building one¶
There are three ways in:
- CSV upload — the usual route for anything above a handful of servers. See Importing from CSV.
- Manual entry — add individual servers. Sensible for small deployments or for filling a gap after an import.
- Location import — bring in an entire site at once, if your organisation uses the Estate module. See Locations and estate.
Hardware configurations¶
Every line is matched to a hardware configuration in our catalogue: one server specification, identified by vendor, model, CPU model, CPU count and memory.
Configurations carry a pre-calculated efficiency grade, and that grade is what every energy figure in your results is built from. Servers sharing a specification point at a single catalogue entry rather than each holding their own copy — which is why correcting a match once fixes it everywhere it appears.
A configuration we have not graded before is sent for grading as soon as you confirm its match. Grading happens in the background, and the line shows a clock in its status column until it lands — hover it for the detail. There is nothing to wait for: a run grades anything still outstanding before it starts, so a pending grade never reaches your results.
Lines matched only approximately are not graded until you confirm them on the review screen, and neither are lines whose match you rejected. Grading a guess would tell you about a server you are about to correct, so we wait for your decision. Nothing is held up by this: a simulation cannot run until its matches are reviewed anyway.
How matching works¶
Matching runs in stages. We check for a known alias first, then an exact match on vendor, model and CPU, then a fuzzy match if neither lands.
The fuzziness is deliberate: inventories rarely record hardware the way vendors name it, and an exact-match-only approach would reject most real data.
Each match carries a confidence score. Lines we are confident about are matched silently and marked validated. Lines we are unsure about are flagged for you to confirm.
A line that differs from our catalogue entry only in how it was written is not
flagged: Dell PowerEdge R7615 / AMD EPYC 9534 is PowerEdge R7615 / AMD Epyc
9534, whatever the score. The vendor repeated in the model, capitals, spacing
and a CPU's clock speed are all ignored. A different suffix is never ignored —
an R740 is not an R740xd.
A CPU is always matched to the model number you wrote: a Gold 6238 is only
ever matched to a Gold 6238, never to a close-sounding neighbour. Notes some
inventory tools add, such as (2.10 GHz / 22-core) or (8 Cores), are
ignored. If we do not hold that model number, the CPU is left for you to pick
rather than guessed.
HPE servers can be written the short way: BL460c G9, DL380 G10 and
Synergy 480 G10 + are read as ProLiant BL460c Gen9, ProLiant DL380 Gen10
and Synergy 480 Gen10 Plus.
Correcting a match¶
One card per distinct entry, not per line. If your file names the same server five hundred times, we matched it once and the review screen asks you about it once — the card says how many lines it stands for, and confirming or skipping settles all of them. A 1,000-line file usually comes down to a few dozen decisions.
Lines are grouped on what you wrote: vendor, model and CPU, ignoring case and extra spaces or tabs. The same server with two different processors is two cards, because they are two different matches.
A decision carries to the other cards. When you confirm a card, the model you confirmed is filled in on every other card in the import with the same vendor and model, and the CPU on every card with the same CPU. A card that then has nothing left to decide settles on its own. A card that still has an open question stays, with the decided field marked As decided on line N, so you can see where it came from and change it if that decision was wrong.
A group covering more than 25 lines says so above the buttons, with the entry it is about to settle. Worth reading before you press it — one click really does settle five hundred rows.
Whatever you decide applies to this import only. Nothing is remembered for the next file you upload.
Open the flagged line. The right-hand column is what we matched it to, and every box in it can be changed.
Click a box — vendor, model, CPU or GPU — and it opens with what we considered for that field, ranked, with a confidence figure against each. Ours is at the top marked our match, so you can always get back to it. Pick one and it fills the box; nothing is decided yet. You can also just type: each box searches the catalogue as you go, and you are free to enter something it does not contain.
Being able to change the boxes separately matters more than it looks. The model
and the CPU are matched independently, and CPU names differ by a single letter —
a 6338 and a 6338N are different parts with different efficiency. Setting the
right model and the right CPU is one decision, made in two boxes.
Confirm Match is what commits, applying whatever the boxes say. Until you press it, nothing about the line has changed.
Typing something we do not hold¶
You are not restricted to the catalogue. If the server you have genuinely is not in it, type it and confirm — the line is saved exactly as you entered it.
The box will tell you, underneath, that the value is not in the supplier
catalogue and what the nearest entry is. Worth reading rather than dismissing: a
PowerEdge R650 and a PowerEdge R650XS are different machines, and if you meant
the one we hold, the efficiency figures for the one we do not will be a rating
requested by name rather than a catalogue match.
What the row colours mean¶
| Green | Your value and ours agree |
| Amber | They differ — either we corrected your input, or you have changed ours |
They compare the two sides of the row, nothing else. Picking a different catalogue entry therefore turns a row amber rather than green, which is correct: it is telling you the line no longer says what your file said. Whether the value is in the catalogue is the separate note underneath the box.
A CPU is compared on its model, not on how it was written. The clock speed,
(R) and (TM), and the words "CPU" and "Processor" are ignored, so
Intel(R) Xeon(R) Gold 6254 CPU @ 3.10GHz is green against Intel Xeon Gold
6254. A different suffix is a different chip: a 6338 against a 6338N stays
amber.
You can also skip a line — useful when the server genuinely is not in the catalogue. Skipped lines are left out of the analysis entirely rather than guessed at.
Confirm the flagged lines before you run
An unconfirmed match still runs, but the result is only as good as the hardware it was matched to. Flagged lines are flagged because we could not tell — a minute spent here is worth more than any other check.
Editing lines¶
Any line can be edited after import: vendor, model, CPU, CPU count, memory, quantity and reference.
A matched line opens on its match. The server, the CPU and (where there is one) the GPU are shown separately. For each you see:
- In your file: what your file said.
- Matched to: the catalogue entry it was matched to. It is shown in red if the catalogue does not hold it.
- How it was decided: matched automatically, confirmed by hand, awaiting review, or carried from line N when a decision made on another line in the same import was applied to this one.
This makes it easy to check someone else's matching. Saving without changing anything leaves the match exactly as it is, so correcting a quantity or a memory figure never undoes a match you confirmed.
To change one, press Change server, Change CPU or Change GPU. Only that part opens, as a box that lists what the matcher considered for your file's text, with the current value marked current, and searches the catalogue as you type. The other parts are left alone. If you enter something the catalogue does not hold, the box says so and suggests the nearest entry. You can still save it, but it will bring no core count or grade. Whatever you change, your file's original text is kept, and Show original input still shows it.
A line that did not match opens with what your file said, ready to correct. When you save, it is matched against the catalogue again.
Lines can be deleted individually, or several at once.
The server list on an EU compliance report behaves the same way, and adds the drive counts and utilisation that the label needs.
Reading the server list¶
Every line shows one status. There are two questions: is its server and CPU in the supplier catalogue, and has it been graded.
| Status | What it means | What to do |
|---|---|---|
| Graded | In the catalogue, and graded | Nothing |
| Grading… | In the catalogue; the grade is on its way | Nothing. Starting a run fetches anything outstanding first |
| Not graded | In the catalogue, but our supplier has no grade for it. Rare | Change the server or CPU to one that is graded, or delete the line |
| Not in catalogue | Did not match, or was matched to a server or CPU the catalogue does not hold. The one it does not hold is shown in red | Edit the line and pick a catalogue entry, or delete it |
| To review | The match is waiting for you in match review | Open Review Matches |
| Skipped | Skipped during match review, and left out of the analysis | Nothing, unless you want it back |
The line above the list counts each status. The filter buttons show one status at a time. To clear out the lines you cannot use, filter to one status, tick the box at the top of the list to select every line in it, and delete. The model and CPU columns show the catalogue's names; Show original input switches them to what your file said.
The core total above the list only counts lines whose CPU has a core count from the catalogue. When some do not, it says how many lines were left out.
Retry matching appears only when some lines failed to match or never finished, for example because the service was unavailable during import. It runs matching again for those lines. Lines that already matched are left as they are.
What if the hardware is not in the catalogue?¶
The catalogue covers a large range of server configurations and is refreshed daily. If something genuinely is not there, get in touch — see Getting help.