Documentation

From download to your first filed report

LinkProven automates latency and throughput testing on your network — configurable sweeps, multi-client load tests, or a single-PC run — and writes a ready-to-file Word / Excel / PDF report from every test. This guide takes you the whole way.

01 Install

LinkProven ships as a self-contained Windows build — no Python or other dependency to install. It carries its own runtime plus fping and iperf3. Install it on both laptops.

  1. Open the Download page and get the installer (LinkProven-…-setup.exe), or the portable .zip if you would rather not install anything.
  2. Installer: run it and follow the steps; LinkProven goes to Program Files with a Start-menu shortcut. Portable: right-click the .zip → Extract All… to a folder you can write to, then run LinkProven.exe.
  3. Accept the UAC prompt. A short welcome tour opens on the first launch.

SmartScreen warning? The build is unsigned, so Windows may show a "Windows protected your PC" prompt on first run. Click More info → Run anyway. To be sure you have the genuine file, compare its SHA-256 with the checksum on the Download page (also published as a .sha256 file with each release).

Administrator. LinkProven needs to run elevated — it sets static IPs and adds firewall rules through Windows' own netsh, and it asks before every change. It requests elevation (a UAC prompt) when you launch it, and closes if you decline.

02 Connect the two PCs

LinkProven measures the link between two machines: one plays Server, the other Client. Connect them on a dedicated point-to-point LAN.

Server at 192.168.1.1 connected by Ethernet to Client at 192.168.1.2; Client runs iperf3 and fping each case.
The canonical two-machine topology — Server 192.168.1.1, Client 192.168.1.2.
  • Two physical machines: join them with a single Ethernet cable (a crossover cable isn't needed on modern NICs) or through a small switch.
  • Two virtual machines: attach both to the same internal / LAN-segment virtual network — not NAT, not Bridged.

03 IPs & firewall

Each side needs a static IP on the test subnet, and Windows Firewall has to allow the test traffic. LinkProven's Setup tab checks all of this for you and offers one-click fixes — it never changes the host silently.

RoleIPv4Subnet
Server192.168.1.1255.255.255.0
Client192.168.1.2255.255.255.0
  1. Launch LinkProven on each machine. On first launch it auto-detects its role from the bound IP (and falls back to a safe default with a heads-up banner if neither matches).
  2. Open the Setup tab. Any row in yellow or red has a fix button — each one shows the exact netsh command it will run and asks for confirmation:
    • Set the correct IP — assigns the role's canonical static IP.
    • Firewall rules — allows ICMP echo, iperf3 (5201), and the control channel (5202).
    • Disable Wi-Fi — if Wi-Fi is live on the test subnet it can corrupt the measurement; this turns it off and re-enables it when you close the app.
  3. When every row is green on both machines, you're ready to run.

Need a different addressing scheme (e.g. a 10.x.x.x LAN, or a router between the two sides)? The Setup tab has a per-PC custom IP override, and the Config tab lets you change the Server/Client IPs, ports, protocol, payloads, bandwidths, and durations — then save it as a reusable profile.

04 Run a sweep

  1. On the Server, the listener starts automatically — leave the window open. It shows Listening on 192.168.1.1:5202.
  2. On the Client, open the Single-Client tab. You'll see the case grid (payload × bandwidth, 20 rows by default), each row ticked. Keep them all for the full sweep, or untick rows / use the per-payload and per-bandwidth quick toggles to run a subset.
  3. Press Run full sweep. The Client connects to the Server and walks every case in sequence — latency and loss via fping, throughput / jitter / loss via iperf3 — filling in the table live, with charts and a duration-aware ETA.
  4. When it finishes — about 11 minutes for the default 20-case grid, warm-up included, plus about half a minute for each case that is retried — LinkProven saves the report set and shows a summary. Press Stop any time to end early and cleanly.

Working through a campaign

For a whole train set rather than one cable, fill in the Plan page first: the train, contract and procedure, who tests and who witnesses, and the test points (typed, laid out as TP1–TP8 in one click, or imported from a CSV copy of the procedure's results table). Each test point's status — PASS, FAIL, REPEAT, INCOMPLETE or not run — is read from the saved reports, and the page names the next one to run. On Single-Client, picking a test point fills its cable ID and ends, and the finished run's summary sheet offers Next and Retest. With No campaign selected, none of this is in the way.

Continuous (multi-segment) mode runs the same grid across several physical segments in one walk-through — e.g. cab-to-cab on a train — and rolls them all into one consolidated report with cross-segment comparison tables.

05 Read the report

Every sweep writes a self-contained folder under Reports/ with a Word document, an Excel workbook, and a .json sidecar (PDF and TXT optional). The headline is the 8-column performance table; an analysis appendix adds per-metric charts and breakdowns.

A sweep produces a Word document, an Excel workbook, a JSON sidecar, and optional PDF/TXT.
What one sweep produces.
  • Cable verdict — with cable validation on (the default for a two-laptop sweep), the report opens with a Table-2 row — Test point · Cable · Loss % · Pass / Fail · Comment — followed by the link and laptop evidence behind it. See Judge a cable.
  • Performance Metrics table — one row per case: payload, bandwidth pushed, throughput received, jitter, loss, and min / avg / max latency.
  • Per-case detail — status, return codes, error notes, and the exact iperf3 / fping command used for that case.
  • Title block — run ID, timestamps, both IPs, tool versions, plus any test-record metadata you filled in (technician, customer, hardware serial, environment, record ID).
  • Analysis tab — load past runs back in to overlay, compare, and diff them, then export a comparison report. Right-click a run to replay its exact plan.
  • TCP runs — show the sending laptop’s retransmits and TCP round trip in place of jitter and loss. DSCP marking (Config) marks the test traffic for the sweep and the report says whether the mark was seen leaving the laptop.

See a full 20-case example: sample report (PDF) · Word (.docx).

06 Load one server with many clients

The sweep answers “how does this link behave across the grid?”. The Multi-Client tab answers a different question: does a server still hold up when a floor of workstations hits it at once? It runs one manually parameterised case — payload, bandwidth, duration and a number of clients — driving all of them from this machine as parallel streams.

It is client-only: there is no LinkProven on the far end and no role gating, so it works in the Server, Client and Loopback roles alike. The page follows the order you actually work in:

  • Parameters — server host and port, client count, payload, bandwidth, duration, plus any extra hosts to ping alongside the throughput test.
  • Can this machine make that load? — a capacity budget that turns your parameters into the real wire rate (payload plus 66 bytes of framing, a 33 % surcharge at a 200-byte payload) and checks it against the NIC’s link speed. 30 clients × 30 Mbps at 200 bytes is 1.2 Gbps on the wire, so over 1 GbE the run is blocked before it spends 30 seconds producing a meaningless number. A NIC that reports no link speed never blocks.
  • Commands to run on the server — pick the server’s Linux distribution and the tab prints the exact install, firewall and iperf3 -s lines to paste into a root shell. The firewall block opens both TCP and UDP: iperf3’s control connection is TCP on the same port the UDP data uses, and a UDP-only rule fails in a way that reads exactly like a dead server.
  • Preflight — pings the host, checks the port, and completes a real iperf3 handshake before you commit to a run.
  • Live log — iperf3’s per-second row for each simulated client and a [SUM] line for the whole load, with one [fping] line per second carrying that second’s reply count and min/avg/max round trip — so a latency spike can be read against the throughput from the same second.

Loopback cannot run a multi-client UDP test. The bundled Windows/Cygwin iperf3 server fails on -u -P N for N > 1 — a known Windows-receiver defect upstream. Against the Linux server this is built for it works normally, and LinkProven says so up front rather than letting you find out through a failed run.

07 Turn results into pass / fail

By default a report states what happened. Fill in Reference values on the Config tab and it also states whether that was acceptable.

Set a limit for throughput, jitter, packet loss, or min / avg / max latency. Throughput also supports a percent-of-offered mode, which is usually what you want: one absolute number is meaningless across a sweep that offers 10 Mbps in one case and 90 Mbps in another, whereas “at least 95 % of what was offered” holds at every step.

Every saved report then gains a verdict banner (how many cases passed, were marginal, or failed), the reference values printed beside each measured value with breaching rows highlighted, the reference table itself so the report records the yardstick it was judged by, and an exceptions table listing only the cases that breached it.

  • A report holds its reference by value. Editing the thresholds later cannot retroactively re-judge a report you already saved.
  • In the Analysis tab, runs judged by the same reference get a threshold line drawn across the throughput charts. Load runs with different thresholds and the line is omitted — comparing them against a single line would be a lie.
  • Arm no thresholds and every report is byte-for-byte what it was before.

08 Judge a cable (cable validation)

With Judge the cable (PASS / FAIL) ticked — the default since 0.7.4 — every two-laptop Single-Client report leads with a verdict for the cable under test: PASS, FAIL, REPEAT or INCOMPLETE. The settings sit on the Config tab in Cable validation (Single-Client sweep); press Apply after a change. This section is the settings reference — Cable validation explains how the verdict is reached.

Off means the old reports. Untick Judge the cable for a plain measurement report: no laptop snapshots, NIC counters, warm-up or retries are recorded, and every format is what it was before 0.7. A Loopback (single-PC) run is never judged — there is no cable — and neither is a TCP sweep.

The criteria

SettingDefaultWhat it does
ScopeAll casesOr Cases up to … Mb/s: only cases at or below that bandwidth are judged (50 = the 12-case scheme); the rest are measured and reported for information.
Expected link (Mb/s)100The link speed both network adapters must report. 0 = any speed. 100 suits a two-pair cable.
Require full duplexOnBoth adapters must report full duplex.
Throughput ≥ (% of expected)99Share of what each case can deliver: its pushed rate, or the link's payload capacity when the case asks for more than the wire can carry (marked ▲, judged on UDP loss only).
UDP loss ≤ (%)0UDP loss allowed in any one judged case.
…or no worse than the bench baselineOnAlso accept a case that lost no more than the bench baseline lost in the same case. Set the baseline on Single-Client → Validation → Use last run or Load from report…
Ping loss ≤ (%)0Ping loss allowed over the judged cases.
NIC error counters must not riseOnFail the cable if either laptop's adapter counts a receive or transmit error — damaged frames.

Warm-up and retries

SettingDefaultWhat it does
Warm up first (not recorded)OnShort settle runs before the sweep, repeated until the laptops lose nothing and neither adapter counts an error.
Warm-up run200 B @ 50 Mb/sThe payload and bandwidth of each settle run.
Attempts / seconds3 / 20How many settle runs at most, and how long each one lasts.
Also ping ≤ (ms)offOptional extra rule: the warm-up's average ping must be at or below this. Settled laptops on good cables averaged 1.5–2.5 ms at 200 B / 50 Mb/s, so keep any limit loose.
Stop if they never settleOnStop the sweep rather than spend ten minutes on a run you cannot use. Off runs it anyway and the report says the laptops never settled.
Re-run a lossy case, up to2Extra attempts for a case that lost packets while neither adapter counted an error and both cards' counts agree. It stops at the first clean attempt; every attempt stays in the report and the last one is judged.

Before and during the run

  • Laptop readiness (Setup tab) — run it on both laptops before the bench run and before every test point. Apply all fixes… switches to Best performance and turns off the adapter's power-saving features with a single adapter restart, after you confirm the list.
  • Test point and Cable ID sit beside Cable length on the Single-Client tab and fill the report's Table-2 row. Run sweeps back-to-back numbers each segment for you and writes one row per test point.
  • The Validation sub-tab colours a payload × bandwidth grid as each case finishes — green no loss, red loss, amber retried or lost inside a laptop, grey not run yet — beside packets lost per second, both laptops' NIC error tiles, the warm-up state and the verdict so far.

09 UDP buffer, fping timeout and priority

Three Config → Test plan settings came out of bench runs (see two lessons from the bench). They apply to every sweep, validated or not.

SettingDefaultWhy
UDP buffer (iperf3 -w, KB)2048 (2 MB)Windows' default 64 KB holds only ~10 ms of 200 B traffic, so a short pause on the receiving laptop drops datagrams its network card already received. The Client passes -w to the Server, so both laptops' buffers grow; the report shows the buffer the Server actually got. It is also the Client's send queue: in a case that fills the link, pings wait behind it (up to ~0.2 s). UDP only. 0 = iperf3's default.
fping Timeout (-t, ms)1000How long fping waits for each reply. Without it, fping in count mode waits only the -p period — 10 ms — so every slower reply used to be reported as a lost ping. Applies to the sweep and the Multi-Client tab; a -t in fping Extra args wins. 0 = fping's own default.
Run iperf3 / fping at high priorityOnStarts both at Windows HIGH priority, so background work (Defender, Windows Update, indexing) pauses them less often — a pause on the receiving laptop is what loses packets at 90 Mb/s. Each laptop applies its own setting.

10 Try it on one PC (Loopback)

No second machine handy? Switch the role to Loopback on the Setup tab and LinkProven runs both ends on 127.0.0.1. It's the quickest way to see the full workflow and a real report without any cabling or IP setup.

Loopback mode runs both Server and Client roles on 127.0.0.1 on a single PC.
Loopback mode — both roles on a single machine.

11 Troubleshooting

SymptomFix
Setup tab: Local NIC IP is red (FAIL) The static IP isn't set or the NIC is on the wrong subnet. Click the row's Set the correct IP button (it falls back to opening Windows adapter settings if it can't detect the adapter).
Firewall rows stay yellow after pressing Fix LinkProven isn't elevated. Close it and relaunch — accept the UAC prompt so it can add the netsh rules.
Orange banner: "IP doesn't match the saved role" The bound IP no longer matches the role (NIC reset, DHCP, moved LAN). LinkProven never flips the role behind your back — either click Set the correct IP, or change the role on the Setup tab.
Phantom packet loss in the report Wi-Fi is live on the test subnet and Windows is routing some packets out the wrong NIC. Use the Setup tab's Disable Wi-Fi fix (it's re-enabled when you close the app).
Loss at 70–90 Mb/s Usually the receiving laptop, not the cable. Check the report's Cable delivered column and its evidence section: if the Server's card counted every frame the Client's card sent and neither card counted an error, the datagrams were dropped inside the laptop and the verdict is REPEAT, not FAIL. Run Laptop readiness on both laptops (mains power, Best performance, adapter power-saving off), keep the UDP buffer at 2048 KB and high priority on, and run again. A short card count or any NIC error is the cable.
Lost pings but no lost datagrams, maximum ping just under 10 ms The signature of fping's 10 ms count-mode wait in reports made before 0.7.3. Keep fping Timeout at 1000 ms and re-run; slow replies are then slow pings, not lost ones.
The sweep stops before the first case: the laptops never settled The warm-up lost packets or counted NIC errors on every attempt. Run Laptop readiness on both laptops, give Windows a minute after start-up, and try again — or untick Stop if they never settle to run anyway and have the report say so.
A 30 s case takes about 31.5 s Expected — about 1.5 s of each case goes to starting the tools, the control-channel handshake and fping's last replies. The progress bar and ETA already account for it.
Sweep finished but no files in Reports/ Auto-save is off (the default) and you clicked Skip on the save dialog. Click Save report now on the Save Options tab — the result stays in memory until the next sweep starts.

12 FAQ

Does LinkProven phone home or collect any data?

No. It runs entirely on your local network. The only outbound request it makes on its own is a check for its own updates against GitHub Releases when it starts — turn off Check automatically on launch on the About page and it makes none. No analytics, no accounts. See the privacy policy.

Why does it need Administrator?

To set static IPs and add Windows Firewall rules for the test traffic, via Windows' own netsh. It always shows the exact command and asks before making a change, and when you close it, it puts an address it set back to DHCP (one you typed yourself stays).

Is it free? Is it open source?

It's free to use for your own network testing, but it is not open source — it's released under a proprietary license (see LICENSE and the terms), which permits use but not redistribution or modification. The bundled fping, iperf3 and other components keep their own open-source licenses.

How do I update?

Open the About tab → Check for updates. LinkProven downloads the new build from GitHub Releases, verifies its SHA-256, swaps the install, and relaunches. Your settings and saved reports are preserved.

Can I change the test grid?

Yes — the Config tab is a full test-plan editor. Override payloads, bandwidths, duration, protocol, IPs, ports, and fping flags, and save named .json profiles.

Still stuck? Write to support@linkproven.com; the log, %APPDATA%\LinkProven\logs\linkproven.log, helps.