i9x Performance

User manual

How to load a results file, read what it says, compare runs, present it to the business and launch a load test — and what your plan includes and consumes.

Choose Save as PDF as the destination in the print dialog.

1. What i9x does

i9x reads the results file a load test leaves behind — the JTL/CSV that JMeter writes — and answers what the default JMeter report does not: where the percentiles really sit, at which point the system saturates, which errors are functional noise, and what conclusion you can put in front of someone who was not in the room.

It also has its own load engine, so you can launch a test and watch it arrive instead of running it elsewhere and importing the file afterwards, and two review pages that turn the measurements into a go-live decision for the business.

The same product is sold in two editions, and you can tell them apart by how you got in:

Edition How it works
Desktop Installed on your machine. You sign in with the same account as the web portal. Your plan comes from the license installed on the machine, and reading results is recorded as activity in your account, like online, so it needs a connection. Credits never block the application (section 11).
Online Opened from a URL. Your plan comes from your subscription, reading results is recorded as credits, an activity indicator (section 11), and your invoices and cancellation live in the web portal.

Your results file never travels. The file you load is parsed in the browser itself. Nothing is uploaded to be analysed, in either edition. Online, the only thing the server is told is how many rows you are reading, so it can record your activity.

2. Signing in and the screen

Getting in

  1. Sign in with the email and password of your i9x account — the same ones you use in the web portal, in both editions. Remember my email fills the address in next time. If your account has a temporary password you are asked to choose a new one first: at least 12 characters, mixing letters and numbers.
  2. Accept the three documents: the Terms and Conditions, the Plans, Credits and Billing Agreement, and the Privacy Notice. Each one has a Read button that opens the full text — the same text published in the web portal. You see this step again only when a new version of the documents is published.
  3. License. During the 14-day trial you go straight in. Starter, Pro, Business and Enterprise have no trial: paste your license key, the long key starting with IQ9- that arrives in the email "i9x Performance - your license and receipt". You can also copy it from My account → License in the web portal. When your month or year ends, the renewal brings a new license and the dashboard asks for it.

Forgot my password, on the sign-in screen, sends a recovery link to your email; it needs an internet connection and the link expires in one hour. Changing your password is done in the web portal.

The trial is counted from the day the account was created, so reinstalling the desktop app does not start it again.

One session at a time. On Trial, Starter, Pro and Business an account can have one dashboard open. Signing in on another machine or window closes the previous session, which goes back to the sign-in screen within a minute. Enterprise has no such limit.

The screen

The window has these areas, top to bottom:

Menu What it holds
My profile Downloads, usage and limits, subscription, renewals, license key and Sign out (section 11).
Analysis Everything you do with the loaded test, in three groups. View: Summary, the entry view, and Show runner, which opens the runner and then reads Hide runner to close it. Reports: Business review, Business continuity and Business context (section 8). History: My runs and Save current analysis (section 7).
Help Light or dark theme, language (ES / EN, applied to the whole interface at once), this manual, Tour dashboard (a guided walk through the screen in the order you work with it), What is i9x?, Privacy (the full privacy notice you accepted) and About i9x.

A strip at the very top tells you how many days are left in your trial, or when your license expires. Its × hides it until the next sign-in; a problem with the license itself cannot be dismissed.

3. Loading your results

  1. In the sidebar, under Upload files, press Choose files.
  2. Pick one or more .csv / JTL files. Several at a time is the normal case: each one becomes a dataset you can select, rename or remove.
  3. If you already had files loaded, you are asked whether to replace them or add to them. The On upload selector remembers your answer if you always want the same one.

When you add, no sample is counted twice: a file with exactly the same content as one already loaded is skipped; a file with the same name that has grown (a JTL JMeter is still writing) only adds its new rows; and a different run with the same name, such as two results.jtl, is added apart as results (2).jtl. A message says which of the three happened.

What the load does with damaged rows. Nothing is fixed in silence. A row without a readable timestamp or without a valid elapsed time is left out (it cannot be placed in the test), and so is the last line of a file that was still being written. Rows with missing or extra columns, a file without a header row, unreadable Latency or Connect values and missing columns are loaded and repaired. Both cases are listed, with counts and file names, in the Data quality card under the KPI bar.

Each file you load is a new analysis and counts as credits for its rows. On a paid plan credits never stop a file from loading. On the trial, what is checked before the rows load is its limit of rows per analysis and its quota: if the file does not fit, it is not loaded and a dialog explains why and what you can do. Files that were already loaded stay where they were. See section 11 for how it is counted.

Two settings that change what you see

Setting What it decides
Time origin Each file starts at 0 overlays runs that happened on different days, which is what you want when comparing. Shared clock keeps the real timestamps and places them chronologically, which is what you want when the files are pieces of the same run.
Source units Whether the file's times are milliseconds or seconds. Leave it on Auto-detect unless the numbers look a thousand times off.

If you see "Showing X of Y records". The loaded datasets together passed the memory limit of the browser, and the rest was dropped so the page would not run out of memory. The analysis is still valid for the rows shown, but it is a sample: remove other datasets before you report conclusions from it.

4. Filtering what you analyse

Every filter in the sidebar applies to every view at once, so a number in one tab always matches the same number in another. The one exception is the one you decide yourself: Exclude results leaves samples out only in the views switched on in its Apply in list (see Exclusions below).

Global filters

Advanced filters

Available from the Pro plan upwards.

Global thresholds (OK | WARNING status)

Its own card, from the Pro plan upwards, laid out like Exclude results and with no buttons: every change applies as you type. Everything starts switched off.

Exclusions

Also from the Pro plan upwards.

Exclusion TABs

Decides which views are computed when a file loads. Every view except Summary starts switched off: tick the ones you are going to open. Each view left off is one less thing to compute, and on a large file that is the difference between a few seconds and a wait. Summary is always on.

5. The views, one by one

Every view opens with its predesigned charts: they are already built, so you only have to read them. The View selector at the top of each view moves between its sections (by time, by label, by virtual users...). Each chart can be shown or hidden and has its own reading (View analysis), its data table and a CSV export.

Advanced analysis carries the same padlock in every view, from the Pro plan upwards: the Advanced Analyst section of each view, the Advanced analysis group at the end of the selector (the readings that have no predesigned chart, such as the degradation and risk indexes, the correlation map or the error Pareto), the Chart builder, custom KPIs, and customizing or importing a chart.

View What it answers From plan
Summary The front page and the view you land on: a health verdict for the test, eight headline KPIs and six charts — throughput, percentile and average of connect / latency / elapsed, errors and error rate by API, the 2xx / 4xx / 5xx split by API, and the 3xx / 4xx / 5xx codes. Start here. Trial
Latency How long requests took — elapsed, time to first byte (latency) and connect, on average and at your chosen percentile — over time, by label and against active users. The table by label stays below the charts. Trial
Errors What failed, when, and where it concentrates: an overview, then by response code, by error message, by label and by domain. Trial
Table Analysis Results grouped by API: samples, success %, transactions per second, averages for elapsed, latency and connect, and the OK/WARNING/CRITICAL status against your thresholds. Above the tables, the labels ranked from slowest to fastest. Trial
Throughput Transactions per second over time, by label and against active users. This is where the capacity ceiling shows up: the point where more users stop adding throughput and only add latency. Starter
Samples How many samples there were and how they were spread: over time, by label and by active users. Pro
Degradation Where response times, throughput and errors start to come apart, over time and as the load grows. Pro
Trend Where the figures were heading during the run: response time, percentiles, connect and active users, and each label over time. Pro
Modal Analysis The shape of the distribution. Two peaks usually mean two different behaviours hiding behind one average — a cache hit and a cache miss, for instance. Business
Complements Secondary readings that do not belong to any single chart above. Business
Thread delay The wait between the steps of each thread, by type of wait, by thread, by label and over time: whether the generator, not the system under test, is the one holding things up. Business
Gantt The transitions between labels and their idle time. The timeline diagram is in Advanced analysis. Business
Compare Baseline against candidate, with a verdict and charts by run. See section 7. Enterprise
Chart builder Your own charts: pick what the chart is about and combine its metrics, statistics and groupings. Every figure follows the same calculation as the rest of the dashboard and the global filters apply. Pro

Every chart shows the figures behind each series — its value over the whole test and its maximum — so you do not have to read values off the axis.

Whatever your plan does not cover stays visible with a padlock: a view in the tab bar, an option of a selector, a button. Clicking it explains what the feature does and which plan includes it; nothing is hidden from you by surprise. A view that is covered but stays empty is usually switched off in Exclusion TABs.

6. Reading a chart

Every predesigned chart has a View analysis button. It opens a panel with three things:

The red lines: degradation and saturation

Charts over time can carry two red vertical lines, in every plan:

A line is drawn only when the data shows it: a run without a ramp of users has no saturation point, and a stable run has no degradation. View analysis explains each line with its figures, or says why there is none. They also appear in Summary and in the Throughput and Degradation views of Advanced analysis.

The DEGRADATION and SATURATION switches in the KPI bar show or hide each line on every chart at once; they only change what is drawn, never a figure. The Summary of tests cards Degradation and Saturation give the clock time of each mark, how far into the run it falls and the users at that moment, for the Total and for the Filtered data, both with the filter percentile used by the red lines. Without a mark they say why: not detected, not reached, no user ramp or not enough data.

Colours of the HTTP classes

Every chart paints the response classes the same way: 2xx in green, 3xx in orange-red, and 4xx and 5xx in red, 5xx (the server errors) the darkest. Several codes of one class get shades of the same colour, and the class is always written in the legend, the tooltip and the table, so you never depend on the colour alone.

The views in Advanced analysis keep their View Insights button: a functional and a technical definition, the key findings and recommendations of what to look at next.

Both come from a local, deterministic engine: the same data always produces the same reading, which is what makes it safe to paste into a report.

Customizing a chart

From the Pro plan upwards, Customize on a chart unlocks Edit query: its metrics, statistics, groupings, X axis and filter. Charts are saved on this computer for your account. Export all and Copy JSON share only the chart definitions, never your results; Import brings in the charts a teammate shared with you, and Restore predesigned puts the original charts back.

The cards of the Summary view do not have the button: each one carries its own one-line reading instead.

7. Comparing runs and History

The Compare view

Load at least two datasets, open Compare, and pick a baseline (what you had) and a candidate (what you want to approve).

  1. Each metric shows its delta and its delta percentage. On latency metrics a positive delta means the candidate is worse.
  2. Set the regression thresholds: warn above X %, fail above Y % (10 % and 25 % to start with).
  3. Each row gets a verdict — PASS, WARN or FAIL — and the summary tells you how many metrics crossed each line.
  4. Export comparison (CSV) saves the table as it stands, for a ticket or a release record.

The All datasets matrix compares every loaded file at once, which is the quick way to spot the odd one out in a series of runs.

History: keeping runs to compare later

Compare needs both files loaded. History keeps the result of a run so you can compare against it weeks later without the file.

Saved runs are also what the Business and Technical reviews compare against (section 8). Saving and reading your history does not consume credits.

8. Business and technical reviews

Two full-screen pages that read the loaded test for people who decide rather than measure. They open over the dashboard; Back to dashboard returns you exactly where you were, with your files, filters and tab untouched.

Page What it answers
Analysis → Business review What the test means for the business: a verdict — GO, GO with conditions or NO-GO — the objectives per flow, capacity, the break point and its projection, impact, value at risk, the main culprits, the comparison with a saved run, recommendations, and the scope of the test.
Analysis → Business continuity Can the business keep operating? Peak, full day, recovery, isolation and RTO; what the test does not prove; a scorecard per critical flow; degradation, recovery against the RTO, residual risks and the plan.

Business context

Without context a review can only say what was measured; with it, it can say what that means. Analysis → Business context opens it over the last review you used (Business review the first time); it is the same context for both, and has three tabs:

The context is saved on this device for the set of files loaded. A new set of files starts from the last context you used, without its injected failures. Until you fill it in, the review uses default objectives and says so at the top.

Filters of the review

Each page has its own filters — all flows or critical flows only, percentile, time window, what counts as a failure (everything, or system failures only: 5xx and no response), a from / to range in minutes, and Compare with a saved run. They apply on top of the sidebar filters, and do not change them.

Export to .PPT turns the page into a presentation with native, editable charts, and Include technical annex adds the supporting detail at the end. The export needs the Pro plan or higher.

9. Exports and reports

Output What it is From plan
CSV export The data behind a table, as it is filtered on screen. In the runner, Download CSV saves the samples collected by a run. Trial
Executive Summary (PDF) The conclusions of the run in a document you can send: KPIs, the relevant charts and the automatic findings. Starter
Results report (PDF) The visible charts of a view or of the Chart builder, each one with its reading and its definitions. Starter
Chart definitions (JSON) Export all and Copy JSON save how the charts are built, never your results, so a teammate can import them (importing is from Pro). Trial
Executive Presentation (PPTX) The same material as slides, for when the audience is a meeting rather than an inbox. Pro
Business / Technical review (.pptx) The review pages as a deck, with an optional technical annex. See section 8. Pro
Comparison (CSV) The baseline-against-candidate table of the Compare view. Enterprise

All of them are generated from what you have on screen — filters included. Filter first, export after. Exporting an analysis that is already loaded does not consume credits.

10. Running a load test

Open Analysis → Show runner. The runner has three parts: the API sequence (the endpoints, executed top to bottom), the load scenario (how hard and for how long) and the options (how the APIs share that time).

Start from a test type

If you know what you want to find out but not which method gives it, start from one of the four test types. Each one fills the fields and leaves the method it chose visible, so every field stays editable:

Test type What it does
Linear Uniform growth at a comfortable level. The starting point: it tells you what normal looks like before you push.
Volume Moderate load held for a long window. Finds what only shows up with time: leaks, connection pools, logs filling up.
Stress Climbs to three quarters of the allowance to find the level at which response time degrades.
Spike A short, wide hit entering on a doubling curve. Reproduces a flash sale or a release announcement.

Execution methods

The Definitions button next to the selector explains each one inside the application.

Method Use it when
Closed model — Constant VUs A fixed number of virtual users, each starting its next request when the previous one finishes. The closest to a classic JMeter thread group.
Closed model — Ramp VUs Users grow up to a target. Use it to find the point where the system gives in.
Open model — Arrival rate The default. Requests arrive at a rate regardless of whether the previous ones finished. This is how real traffic behaves.
Classic thread group (JMeter) Reproduces the familiar JMeter semantics.
Total iterations A fixed amount of work rather than a fixed duration.
Exponential / Linear ramp-up Custom growth curves, written as a JSON array.
Local simulation only No traffic leaves the machine: synthetic samples to try the dashboard out. Clearly marked as simulated in the run metadata so it never gets mistaken for a measurement.

The fields, and their rules

Field Rule
VUs Minimum 1, maximum 1200.
Ramp-up Minimum 1, maximum 3600 seconds, and never greater than the duration.
Duration Minimum 1, maximum 3600 seconds.
Time-Wait Pause between iterations for each user. Not greater than the duration.
Timeout How long to wait for a response. 0 disables it.

Ramp-up and Duration follow the Data in selector, so check whether you are typing seconds or milliseconds before launching.

Check it before you press Run

Options

Running it

Run test starts the sequence. Pause test holds it without losing the samples collected (the current API finishes first) and Resume test carries on. Stop test ends it and keeps the data. Download CSV saves the samples once there are some, and Clear test resets the scenario and wipes the runner's dashboard.

With Real time monitoring enabled, samples arrive in the dashboard while the test runs and it redraws at the interval you choose — from one second to thirty minutes, or continuously. Real-time monitoring and sequences of several APIs are included from the Pro plan upwards.

Closing the window or pressing stop ends the run: the engine does not keep generating traffic against someone's API once nobody is watching.

Launching a run is free. Running, repeating or reconfiguring a test never consumes credits. What does, online, is reading its results in the dashboard through monitoring (section 11).

Only run tests against systems you are authorised to test. In the online edition the engine is restricted to an allow-list of targets; a target outside it is rejected before a single request is sent.

11. Plan, credits and account

The My profile menu

Entry What it opens
My details Your name, email, company and plan, and where to change your password. The password is changed in your i9x account, so the new one works here and on the website (it needs an internet connection). Name, company and billing details are edited on the website.
My Downloads Installers and your license certificate.
My space Usage and limits: your plan, your activity this period in credits next to your plan's reference (or, in the trial, the rows and reports used of its quota), what that is equivalent to, when it resets, and the rows your plan allows per analysis.
My Subscribe Your current plan, where it comes from, and how to change it.
My Renewals When you are charged next, or when the license expires.
My license On desktop, where you paste a license key. Online, where you read and copy the key issued to you — the one you use to install the desktop app under the same subscription.
Sign out Closes the session on this machine.

Invoices, billing details and cancelling a renewal live in the web portal, not here, so there is only ever one copy of them. The dialogs link to it when it is reachable, and say so plainly when it is not.

What each plan includes

Feature Trial Starter Pro Business Enterprise
Duration 14 days Monthly or annual Monthly or annual Monthly or annual Monthly or annual
Reference credits per month (never a block) No credits: 20,000 rows and 5 analyses in total 800 2,500 10,000 From 50,000 (by contract)
Rows per analysis 10,000 No limit No limit No limit No limit
Dashboard sessions at a time 1 1 1 1 No limit
Summary, Latency, Errors, Table Analysis, with their predesigned charts ✓ ✓ ✓ ✓ ✓
View analysis of each chart, History, chart definitions in JSON ✓ ✓ ✓ ✓ ✓
Load runs with Constant VUs, Ramp VUs or local simulation, one API ✓ ✓ ✓ ✓ ✓
Throughput — ✓ ✓ ✓ ✓
Executive Summary and Results report in PDF — ✓ ✓ ✓ ✓
Samples, Degradation, Trend — — ✓ ✓ ✓
Advanced analysis: the Advanced Analyst section and the Advanced analysis views of every view, the Chart builder, custom KPIs, and customizing or importing charts — — ✓ ✓ ✓
Business and Technical review pages — — ✓ ✓ ✓
Advanced filters and exclusions, comparing two runs from the History — — ✓ ✓ ✓
Multi-API sequence, every other execution method and custom ramps, real-time monitoring — — ✓ ✓ ✓
Executive Presentation in PowerPoint — — ✓ ✓ ✓
CI/CD quality gate (i9x-gate) — — ✓ ✓ ✓
Modal Analysis, Complements, Thread delay, Gantt — — — ✓ ✓
Compare (baseline vs candidate) — — — — ✓

The trial needs no card, starts the day the account is created, and nothing is charged when it ends: to keep using the dashboard you choose a plan. Paid plans have no trial period.

How credits are counted

Credits are an activity indicator: they measure what you read in the dashboard, not what you run. On a paid plan they never limit, suspend or block the application — going over your plan's reference changes nothing. Two things count:

Unit What it is Cost
Row Each row of results analysed: a row of a CSV/JTL file you load, or a sample of a run you watch in monitoring. 0.15 credits
Analysis Each new analysis you open: a results file you load, or a monitoring session. 5 credits

Example: loading a results file of 10,000 rows is one analysis plus 10,000 rows, 1,505 credits.

What you do Counts?
Load a results file Yes: one analysis plus its rows. Loading the same file again is a new analysis.
Add rows to a file already loaded Yes, the rows added. It stays the same analysis.
Watch a run's results in monitoring Yes: one analysis per monitoring session plus the samples that arrive.
Launch, repeat or configure a run No.
Filter, change tabs, open the reviews, export what is loaded No.
Your saved history, your account and its pages No.

12. When something does not work

What you see What it means
"You used the trial quota" The trial's 20,000 rows or 5 analyses are spent. No more results can be loaded until you choose a plan.
"This result file is larger than your plan allows per analysis" The file has more rows than one analysis of the trial admits (10,000; paid plans have no limit). Nothing was loaded or charged. Split the file or choose a paid plan.
"Your session has ended" It expired, or your account was opened in another session — one session per account on Trial, Starter, Pro and Business. Sign in again.
"Your free period has ended" / "Your license has expired" The trial or the paid period is over. Paste the license of your plan or of your renewal; the screen's I need help tells you where to find it.
"No backend runner detected; using local simulation instead" The load engine did not answer, so the samples you are looking at are synthetic. The run metadata says so. Check that the application is running normally before reading anything into those numbers.
"The load engine rejected the run" (target not allowed) The deployment restricts which systems can be load-tested, and this one is not on the list. That is a configuration decision, not a mistake in your scenario.
"Invalid or missing URL" The endpoint is not a complete URL — the scheme is usually what is missing.
A padlock on a tab That view is not in your plan. The dialog says which one includes it.
A tab that stays empty after loading a file The view is switched off in Exclusion TABs, which is how every view except Summary starts. Tick it there.
"p95 is an estimate in N of M groups" (Data quality) Those labels have fewer than 200 samples, and a p95 needs at least 10 samples above it to stay put. The KPI cards use all the filtered samples and are not affected; the tables by API mark the affected rows with ⚠. Run longer or with more load, choose a grouping with fewer groups, or read a lower percentile (p90 needs 100).
A tab and the KPIs give different figures after excluding results Exclude results is switched off for one of them in Apply in. Switch it on in both, or off in both, to compare them. See section 4.
"Showing X of Y records" The loaded datasets passed the memory limit of the browser. See section 3.
"This data has N labels; the analysis keeps the 25…" The file has more labels (or user levels) than Top N lets through, so the least frequent ones are left out of charts and KPIs. Raise Top N in the sidebar to include them. See section 4.
A page of your account does not open from the desktop app Account pages live in the web portal. The desktop edition works from an installed license, so the dialog tells you where to write instead of sending you to a page that cannot answer.

13. Terms used in this manual

The short list below covers this manual. For every figure the dashboard shows and exactly how i9x computes it, open Help → Glossary of terms or the glossary directly.

Term Meaning
Elapsed Total time of the request, from sending it to receiving the last byte.
Latency Time until the first byte of the response arrives. Elapsed minus latency is roughly how long the body took to transfer.
Connect Time spent establishing the connection. If this grows while latency does not, the problem is usually in the network or the balancer, not the application.
VU Virtual user: one simulated client executing requests one after another.
TPS Transactions per second: completed requests per second.
Percentile (p95, p99) The value below which that share of requests fell. p95 = 2s means 95 % finished in two seconds or less. Averages hide the bad tail; percentiles do not.
Baseline The run you accept as the reference, against which a new one is judged.
Flow A label of the test seen as a business operation — payment, login, sign-up. A critical flow is one the business cannot do without.
RTO Recovery time objective: how long the business has committed it can take to be back after a failure.
Credit The unit that measures your activity: 0.15 per row read, 5 per analysis. An indicator, not a limit — on a paid plan it never blocks anything. Credits have no cash value.
Analysis One results file loaded, or one monitoring session. Each new one counts once.
Trial The 14-day period every new account starts with, with its own fixed quota instead of credits.