How it works

An Ethernet cable acceptance test that can tell the cable from the laptops

LinkProven runs a payload × bandwidth sweep between two Windows laptops across the cable under test, then gives the cable a verdict — PASS, FAIL, REPEAT or INCOMPLETE — in the shape of a test procedure's results table. It records the evidence behind the verdict, including both network cards' packet counts, so a packet dropped inside a laptop is never blamed on the cable.

An illustrative Table-2 row as it appears at the top of a report
Test pointCableLoss %Pass / FailComment
TP3CAB-00420PASSAll four criteria met.

Illustrative. Every report leads with this row, ready to copy into the procedure's results table — in Word, PDF, text and Excel.

01 The test, end to end

LinkProven runs on two Windows laptops joined by the cable under test. One plays Server (192.168.1.1), the other Client (192.168.1.2); the Client drives the whole sweep over a small control channel, so nobody has to touch the Server once it is started.

Server at 192.168.1.1 connected by Ethernet to Client at 192.168.1.2; the Client runs iperf3 and fping for each case.
Server and Client across the cable under test. iperf3 uses port 5201, LinkProven's control channel 5202.
  1. Get both laptops ready. The Setup page checks the IP address, the firewall, Wi-Fi, the power mode and the network card's power-saving features on each laptop, and offers to fix what is wrong — always after you confirm.
  2. Name what you are testing. On the Client, pick the test point from the campaign plan (or type one) and the cable ID. Both go on the report.
  3. Run the sweep. A short warm-up checks that the laptops are settled, then every payload × bandwidth case runs for 30 s: iperf3 sends UDP at the target rate while fping measures latency and loss alongside it. Both network cards' counters are read around every case.
  4. Read the verdict. Each case is coloured as it finishes; the run ends with PASS, FAIL, REPEAT or INCOMPLETE, judged on the four criteria below.
  5. File the report. Word and Excel by default (PDF, TXT and CSV on request), led by the results-table row and the verdict, with every piece of evidence behind it.

02 Why loss alone can't judge a cable

A table of throughput and loss figures tells you what happened, not whether the cable is fit for service. On bench tests of a train's on-board LAN, the same 2 m patch cord lost 6.24 % on one run and nothing on the next. Only the laptops had changed: on battery, power-saving network-adapter features switched on, a small receive buffer, or Windows still busy just after start-up.

Packet loss is only evidence against a cable if both laptops could have received everything. So LinkProven checks the laptops first, then judges the cable on evidence that separates the two:

  • Laptop readiness on the Setup tab — run it on both laptops before the bench run and before every test point. It checks mains power, the Windows power mode, the adapter's Receive Buffers, its energy-saving features (Energy Efficient Ethernet, Ultra Low Power Mode, Green Ethernet …), Speed & Duplex and the link. Apply all fixes… switches to Best performance and applies the adapter changes with a single adapter restart — only after you confirm the exact list.
  • Evidence in every report — both laptops' link, duplex, power and NIC settings at the start of the run, each NIC's error and discard counters over the run, and the receiving end's exact lost / sent counts for every case.

03 The four criteria

The cable passes when all four hold. Each is set on the Config tab under Cable validation, and the values used are stored inside every report they judge — a later change cannot re-judge an old report.

CRITERION 1

Link

Both network adapters report the expected speed, in full duplex.

Default: 100 Mb/s, full duplex — a two-pair cable carries no more
CRITERION 2

Ping loss

No ping lost. A ping counts as lost only if no reply arrives within fping's timeout; a slow reply is a slow ping, reported in the latency columns.

Default: 0 %, with a 1 s reply timeout
CRITERION 3

Throughput & UDP loss

In every judged case the received throughput reaches 99 % of what the case can deliver, and no UDP datagram is lost — or no more than the bench baseline lost in the same case.

Default: ≥ 99 %, 0 % loss
CRITERION 4

NIC errors

Neither laptop's network adapter counted a receive or transmit error. Errors mean frames arrived damaged: the cable or a connector.

Default: on

Scope. Every case is judged by default. To judge only the lighter cases — for example the 12 cases at 10, 30 and 50 Mb/s — choose Cases up to … Mb/s; the rest are still measured and reported for information.

Loss % in the Table-2 row is the number of UDP datagrams lost divided by the number sent, over all judged cases, from the receiving laptop's own counts — weighted by packets, not an average of percentages. A loss below 0.001 % prints as <0.001, never as a 0 beside a FAIL.

04 The four verdicts

PASS

All four criteria hold.

FAIL

At least one criterion failed and the evidence points at the cable. The Comment says which.

REPEAT

Packets were lost, but the run shows the cable is not the cause: in every failing case the two network cards' counts agree and neither adapter counted a damaged frame — or the warm-up never settled and the only failure is loss, with clean NIC counters. Run it again.

INCOMPLETE

Nothing failed, but a criterion had no evidence — for example the Server laptop runs an older LinkProven that does not send its adapter's counters. Shown as "Not decided".

NOT RECORDED never reads as PASS. A criterion without evidence — a link speed that could not be read, a missing card count, a case with no throughput reading or nothing received — is NOT RECORDED, and a run that has one is INCOMPLETE at best. A missing count is "not recorded", never 0. A run stopped part-way is INCOMPLETE too, never a PASS on the cases it managed.

05 Did the cable deliver? The card counts

iperf3 reports how many datagrams it received. That is not the same as how many crossed the cable. LinkProven reads both laptops' network-card packet counters around every case, over the same span on each side, and compares them: the frames one card put on the cable against the frames the other card took off it.

The REPEAT case from the sample case matrix: 200 B at 70 Mb/s, 30 s.

What the cards showWhere the loss happenedVerdict
The Server's card counted every frame the Client's card sent, the Client's card every reply, and neither card counted an error Inside a laptop — the cable delivered every packet REPEAT
The Server's card counted even one frame fewer than the Client's card sent On the cable FAIL
Either card counted a receive or transmit error Damaged frames: the cable or a connector FAIL
A lossy case has no card counts Unproven — REPEAT needs proof, so the loss stands FAIL
  • Client → Server gets no allowance. That direction carries the UDP stream; on the bench the Server's card always counted at least as many frames as the Client's card sent, so one short is the cable's.
  • Server → Client allows 20 frames per case. That direction carries the ping replies, plus multicast and broadcast chatter (name resolution, discovery) that a card filters out before counting. A lost ping is put down to the laptop only when its reply provably reached the Client's card.
  • The check assumes the two laptops are the only devices on the link, as in the procedure. With a switch in between, read the counts as information.

06 Cases the link can't carry

Every packet costs its payload plus 66 bytes on the wire — Ethernet header and checksum, IP and UDP headers, preamble and the inter-frame gap. With 200-byte packets, 90 Mb/s of payload needs about 120 % of a 100 Mb/s link: it cannot fit on any cable. The sending laptop waits instead of dropping, so about 74 Mb/s arrives and nothing is lost. The link's capacity is taken from the slower of the two laptops' link speeds.

Such a case is marked ▲ and judged on UDP loss only — never on throughput or ping. That is why a case can read 73.8 instead of 90 and still pass.

07 Warm-up and retries

The warm-up

Before the sweep, short unrecorded runs — 200 B at 50 Mb/s for 20 s, up to three attempts by default — repeat until the laptops lose nothing and neither adapter counts an error. If they never settle, the sweep stops, so you don't spend ten minutes on a run you cannot use (untick Stop if they never settle to run it anyway; the report then says so).

An extra ping limit (Also ping ≤) is available but off by default: good cables on settled laptops averaged 1.5–2.5 ms at 200 B / 50 Mb/s, so a tight limit would stop sound runs. A warm-up that measures nothing at all — the Server could not bind, the firewall blocks iperf3, the link is down — stops at once and shows the real error.

Retries

A judged case that loses packets with no NIC errors, while the card counts show the cable delivered everything, is run again — up to twice by default, stopping at the first clean attempt. Every attempt stays in the report; the last one is judged, and the Comment says the case was retried.

A case with adapter errors, or whose card counts came up short, is never retried — that is the cable talking. And an earlier attempt the cable did not deliver fails the case even if a retry was clean.

08 Two lessons from the bench

Cable validation was built from real bench runs on a train's on-board LAN. Two of them changed how every LinkProven sweep measures, validated or not.

Bench lesson 1 · v0.7.1

Windows' 64 KB UDP buffer was losing packets inside the Server laptop

Symptom
Two cables that had read 0 % lost datagrams on the next runs.
Evidence
In every lossy case the Server's network card had received every packet the Client's card sent, with no NIC errors. iperf3 never got them: Windows gives it a 64 KB UDP receive buffer — about 10 ms of 200-byte traffic at 50–90 Mb/s — and any longer pause on the receiving laptop overflows it.
Fix
The sweep asks iperf3 for a bigger buffer with -w: Config → Test plan → UDP buffer (iperf3 -w, KB), default 2048 KB (2 MB). The Client passes it to the Server, so the receiving laptop's buffer grows too, and the report's evidence shows the buffer the Server actually got. iperf3 and fping also run at Windows high priority, so Defender, Windows Update and indexing pause them less often.
Trade-off
-w also sizes the Client's send queue. 2 MB absorbs about a fifth of a second of the heaviest case; in a case that fills the link, pings wait behind it, so expect ping times up to about 0.2 s there — a slow ping, not a lost one.
Bench lesson 2 · v0.7.3

fping counted every reply slower than 10 ms as lost

Symptom
Runs that lost no UDP datagram in any of 40 cases still read REPEAT on ping loss alone (about 5 %), and every case's maximum ping sat just under 10 ms.
Evidence
fping's own help gives its reply timeout as 500 ms "except with -l/-c/-C, where it's the -p period". LinkProven runs fping with -c and -p 10, so the wait was 10 ms. At 90 Mb/s a full link queue makes most replies slower than that — and the Client's card had received every one.
Fix
fping now waits 1 s for each reply: -t 1000, Config → Test plan → fping Timeout (-t, ms). Ping loss is near 0, and the maximum ping in the 70–90 Mb/s cases is higher — and true.
Tell-tale
In an older report, a case whose maximum ping sits just under the -p period is this bug's fingerprint.

09 Bench baseline and test points

The bench baseline

Before the real cables, run the sweep with the same two laptops on a short, known-good patch cable, then make it the baseline: Single-Client tab → Validation → Use last run (or Load from report…). With …or no worse than the bench baseline ticked, a case may lose as much as the baseline lost in that same case and still pass. With settled laptops a good bench cable loses nothing, so the baseline is usually 0 % and the rule stays strict. A Loopback run, a run with validation off, an incomplete run or a report older than 0.7 cannot become the baseline.

A whole train in one session

Fill in Test point and Cable ID beside Cable length on the Single-Client tab. For several test points in a row, tick Run sweeps back-to-back and name the first segment (for example TP1). After each one a dialog shows that test point's verdict and offers the next name and cable; Retry re-tests a failed one. The session report opens with a Table-2 holding one row per test point, and each test point can also be saved as its own report — the {tp} filename token (e.g. Test2-{tp}) names them.

10 Run a cable acceptance test

  1. Install LinkProven on both laptops and join them with the cable under test — Server 192.168.1.1, Client 192.168.1.2. The documentation walks through it.
  2. Run Laptop readiness on the Setup tab on both laptops and apply its fixes.
  3. Cable validation is on by default (Config → Cable validation → Judge the cable (PASS / FAIL)). Change the scope or criteria if your procedure differs, then press Apply.
  4. On the Client, fill in Test point and Cable ID and press Run full sweep. The Validation sub-tab colours the case grid as each case finishes and shows the verdict so far. With the default 20-case grid a sweep takes about 11 minutes, warm-up included, plus about half a minute for each retry.
  5. Read the verdict at the top of the report; the Table-2 row is ready to copy.

A Loopback (single-PC) run is never judged — there is no cable — and a TCP sweep isn't either, since it has no datagram counts. Untick Judge the cable for a plain measurement report, exactly as before 0.7.

Questions about a verdict? Write to support@linkproven.com.