flowchart LR L[1 Link<br/>joined, signal, retries] --> A[2 Address<br/>IP, prefix, lease] A --> H[3 Next hop<br/>neighbour, gateway] H --> N[4 Naming<br/>query, answer] N --> T[5 Transport / security<br/>port, timeout, TLS] T --> S[6 Service meaning<br/>protocol, auth, payload]
Lab: Network Diagnostics
Lab: Network Diagnostics
Checkpoint: Trace Ownership
You now know:
- A trace record starts with a stable packet identity and then records each boundary in order.
- Link, route, transport, policy, and receiver failures have different owners.
- IoT traces should include energy effect and freshness effect, not only final arrival.
Lab: Network Diagnostics
Start With the First Broken Boundary
Stop at the First Test That Fails
Picture a greenhouse sensor that joined its local network but never updates the dashboard. A gateway means the device or service that joins the sensor network to another system.
Name the link, address, next hop, gateway, name lookup, protected connection, and service reply. Test each boundary in that order with the same message and stop when one promise fails.
Keep commands, packet evidence, times, expected and observed results, and the first failed boundary. This proves one diagnostic path, not the whole network; the deeper lab builds a reviewable evidence packet and tests alternative causes.
A networking lab is not a hunt for random commands. It is a way to find the first boundary where a message path stops meeting its promise: link joined, address assigned, next hop reachable, name resolved, secure transport established, or service response understood.
Imagine that a greenhouse sensor stopped updating at 09:14. “The cloud is down” is only a guess. A useful investigation asks one smaller question at a time and keeps the result of each test. When the first boundary fails, stop climbing and name what the evidence supports.
This lab continues the message-path model from Devices, Packets, and Layers. You will follow one controlled browser scenario and finish with a diagnostic packet that another person could review without repeating your guesses.
Lab Contract and Evidence Ladder
Before you start: Chapter 2 is the only prerequisite. You do not need previous command-line, packet-capture, Python, or hardware experience. The required browser investigation takes about 45 minutes. Hardware, subnetting, service-scanning, and radio-measurement extensions appear later, after their prerequisite chapters.
Your task is to climb six evidence boundaries in order:
| Boundary | Evidence to collect | What passing it does not prove |
|---|---|---|
| 1 Link | Association or cable state, signal observation, retries or loss | A usable address or stable end-to-end path |
| 2 Address | IP address, prefix or mask, lease state, configured gateway and resolver | That the gateway, resolver, or service will respond |
| 3 Next hop | ARP/ND observation and a controlled gateway test | Internet access, healthy radio under every condition, or DNS success |
| 4 Naming | Expected query, resolver used, returned answer or failure | That the returned host accepts a connection |
| 5 Transport / security | Refusal, timeout, connection, or TLS result | That credentials, topic, route, command, or payload are accepted |
| 6 Service meaning | Protocol response, authorisation result, and semantic outcome | That every future request or every device will behave identically |
A single successful gateway ping is evidence that some local path worked during that test. It does not rule out weak or intermittent radio conditions. Likewise, an open TCP connection is not proof that the expected application service is healthy.
Guided Investigation: One Symptom, Six Boundaries
Use the simulator as an evidence notebook, not as a guessing game. Choose Gateway backlog, then collect observations before opening the Fix Plan. For every probe, map the result to one of the six boundaries above.
Work through this sequence:
- State the symptom without naming a cause. Record the device, time, expected behaviour, and observed behaviour.
- Start at the lowest unproved boundary. Do not jump to cloud, credentials, or firmware because those explanations sound familiar.
- Record the observation, not only the test name. “Packet capture” is an action; “LAN packets reach the gateway while WAN acknowledgements disappear” is evidence.
- Name what the observation supports. A result can reduce uncertainty without proving a single root cause.
- Stop at the first failed boundary. Later boundaries are untested unless the evidence genuinely crosses them.
- Choose one bounded next test or owner. Do not change several variables at once.
| Observation | Boundary | Supported conclusion | Still unproved |
|---|---|---|---|
| Device remains joined and nearby nodes report normal link behaviour | Link | A site-wide link loss is less likely during the observation | This device’s address, gateway path, and later boundaries |
| The device has its expected address and gateway configuration | Address | A usable configuration is present | That the next hop responds or forwards traffic |
| LAN packets reach the gateway, but expected WAN acknowledgements disappear | Next hop / transport boundary | The local path reaches the gateway; failure is beyond the device link | Whether the remote service is healthy or the gateway forwards correctly |
| Gateway log shows queue growth and repeated WAN reconnect events | Transport / security | The gateway-to-upstream session is the first observed failed boundary | Service processing after a successful session |
The simulator covers several IoT fault scopes, so its labels are suspect scopes, not automatic proof. Your six-boundary packet is the authoritative interpretation for this networking lab.
Practitioner: Use a Proof Ledger
For each probe, write four short fields: observation, boundary, supported conclusion, and still unproved. This prevents a common troubleshooting error: turning a plausible explanation into a confirmed cause before the path has been measured.
When evidence is mixed, keep the claim narrow. “Eight of ten gateway probes returned during this one-minute test” is defensible. “The radio is healthy” is broader than the evidence.
Under the Hood: Negative Evidence Has Limits
A timeout is not a protocol verdict by itself. It may reflect loss, filtering, congestion, an unreachable route, an overloaded peer, or a timeout that is too short. A refusal is more specific: the attempt elicited a rejection from the endpoint or an intermediary, but it does not by itself identify which one or prove the intended service is present. A TLS alert shows that the exchange reached security negotiation; it is different from an application response such as HTTP 401 Unauthorized.
The first failed boundary is therefore the earliest boundary with evidence of failure, not the earliest place where no success message happens to appear.
Build the Diagnostic Packet
The packet is the lab output. Keep it short enough to complete during the incident and precise enough for another person to challenge. Read the fields as one evidence chain rather than a form to fill in at random. Establish scope and symptom first, then move from link and address evidence through the next hop, naming, transport, security, and service response. Finish by naming the first failed boundary, the next owner or bounded test, and one repeated observation after the change. That order prevents a plausible application theory from outrunning lower-layer evidence.
| Field | Required record |
|---|---|
| Scope | Device or gateway, timestamp, software version, intended endpoint |
| Symptom | Expected behaviour and observed behaviour, without a guessed cause |
| Link | Association or link state, signal/retry observation |
| Address | IP, prefix/mask, gateway, resolver source |
| Next hop | ARP/ND or controlled gateway result |
| Naming | Queried name, resolver, answer or failure |
| Transport/security | Host, port, refusal/timeout/connection/TLS evidence |
| Service | Protocol response, authorisation, payload, or “not tested” |
| Decision | First failed boundary, evidence supporting it, next owner or bounded test |
| Retest | One repeated observation after the change or escalation |
Practitioner: Worked Packet
At 09:14, greenhouse gateway gw-02 is running build 4.7.2 and should reach broker.lab.example, but its queue age rises while local devices continue sampling. Local associations remain stable, and the expected gateway address, prefix, upstream gateway, and resolver are present. Device traffic reaches gw-02, and a controlled lookup returns the expected lab address, so the evidence has crossed link, address, next-hop, and naming boundaries.
The WAN session then shows repeated reconnects while expected acknowledgements disappear. The service is deliberately left untested until a stable upstream session exists, making transport/security the first failed boundary. The next bounded action is to inspect the gateway WAN path and its session logs. After that path is restored, the retest must show acknowledgements resuming and queue age falling. This order demonstrates how the packet stops at the first unsupported boundary instead of decorating an application guess with unrelated checks.
Copyable Blank Packet
Scope:
Symptom:
Link evidence:
Address evidence:
Next-hop evidence:
Naming evidence:
Transport/security evidence:
Service evidence:
First failed boundary:
What remains unproved:
Next bounded test or owner:
Retest result:
Exit Evidence and Continue
Before leaving the lab, make sure you can show all four exit items:
- one completed diagnostic packet;
- one sentence naming the first failed boundary;
- one sentence stating what remains unproved;
- one retest result or a bounded next test with a named owner.
The central habit is simple: do not climb past a failed boundary, and do not claim more than the observation proves. This turns troubleshooting from a list of guesses into a reviewable chain of evidence. Later in Core Networking, companion labs apply this same packet to addressing and neighbour evidence, authorised service reachability, and Wi-Fi link-margin measurements after those concepts have been taught.