Start with a flow, not a vendor
Record the source, destination, protocol, port and incident window. Include the source site, routing context and any address translation relevant to the connection. When a user reports that “the network is slow,” first establish which transaction is affected and whether the issue can be reproduced.
Use a working comparison where available: another site, another application or an earlier known-good period. A comparison is helpful only if the relevant conditions match. Different DNS answers, access policies or return paths can make two apparently identical tests exercise different services.
Evidence to compare across network boundaries
| Boundary | Useful evidence | Interpretation caution |
|---|---|---|
| Access and switching | Interface status, error counters, VLAN membership and relevant neighbor information. | An interface that is up can still experience loss or forwarding problems. |
| Routing | Selected route, next-hop reachability, routing context and relevant adjacencies. | A route entry alone does not prove successful end-to-end forwarding. |
| Security and NAT | Policy decision, session details, translation and matching traffic logs. | Check the actual flow and context; a rule name alone is insufficient. |
| Application edge | Listener or service health and a representative application transaction. | A successful ping does not prove that the required application port works. |
| Return path | Available routing and policy evidence in the reverse direction. | Forward-path success does not establish symmetric return reachability. |
Normalize the meaning while preserving the original evidence
Use a common investigation record for device identity, OS version, timestamp, command or API request and observation. Keep original outputs available for review. Normalizing status labels makes evidence easier to compare, but should not erase vendor-specific semantics.
For example, an established routing adjacency and a permitted firewall session answer different questions. Presenting both as “healthy” can hide an important gap in the investigation. Explain what each check demonstrates and what it does not establish.
If evidence comes from different times, state the collection interval. A counter sampled after service recovery may not describe the incident itself. Missing credentials or an unsupported command should become an explicit gap, not a positive health result.
Illustrative workflow across a branch and data center
Consider a branch router, a data-center switch and a firewall on an application path. The goal is to identify where evidence first diverges from the expected behavior, not to assume that the most recent alert identifies the root cause.
Confirm the affected flow, check the branch route and interface evidence, then inspect the relevant data-center forwarding context and firewall observations. Compare those findings with a working flow where appropriate. If the application still fails while the inspected network evidence is consistent, collect application-side evidence or escalate with the precise checks already completed.
The handover should contain the tested flow, timestamps, observations, exclusions and the next unresolved question. This gives another team a reproducible starting point rather than a generic “network looks fine” message.
Apply the method with NetPrompt AI
NetPrompt AI provides multi-device investigation and configurable workflows across supported network platforms. Use the recorded multi-device demonstration to review the interaction, then validate the actual vendors and OS versions in your estate. A registry entry indicates a driver profile; it is not evidence that every device function or firmware release has equivalent validation.
Combine configured device access with monitoring and ITSM connectors to keep findings in one investigation workflow. Scope permissions to the approved task and preserve human approval for remediation. Start with a small representative device set and test how the workflow handles an unreachable device or an incomplete response.
Common questions
- Does multi-vendor support mean every command is supported?
- No. Validate the required command families, device models, OS versions and permission levels. Request a demonstration of the specific operational task, not only a list of vendor logos.
- Is a topology diagram enough to diagnose an incident?
- A diagram helps establish intended relationships. Diagnosis also needs current forwarding, policy and service evidence for the affected flow.
- Can AI identify the fault from one alert?
- Sometimes one alert is highly informative, but a causal conclusion still needs sufficient context. A useful workflow records the uncertainty and requests the next relevant check.
Continue exploring
- Watch multi-device troubleshooting
- Check the platform registry
- AI troubleshooting workflow
- Create a network RCA
- Add NetBox inventory context
- Turn engineering experience into network troubleshooting context
- Troubleshoot application connectivity across network and servers
- Investigate VPN tunnel failures and branch connectivity
Further technical reading
External references explain general practices; they are not endorsements of NetPrompt AI.