Choose the observation point before capturing
Tcpdump captures packets visible at a selected interface and displays protocol information. It can help investigate whether a connection attempt was sent, whether a response appeared and whether byte ranges were retransmitted during a slowdown. It does not automatically expose every flow on a host through one observation point. Loopback, container network namespaces, tunnels and virtual interfaces can show different traffic. Establish the relevant address, port, protocol and time window first. The capture should answer a specific question rather than accumulate unrelated communications. Observation scope is part of the evidence because a missing packet means something different when the selected interface could never have seen that flow.
Capture a bounded local flow
The example targets TCP port 8080 on Linux loopback for a local test service. The -nn option avoids address and port-name lookups, while -c 20 limits the count. An outer timeout bounds the session even when no matching flow arrives and the count is never reached. Verify actual interface names, capture permissions and installed-version support. The command demonstrates a procedure; no traffic output is presented as a measurement here. If saving to a file, choose retention and size limits and decide whether a restricted header length is sufficient. Count bounds and time bounds solve different problems and are useful together for a focused diagnostic session.
timeout 10s tcpdump -i lo -nn -c 20 'tcp port 8080'Read connection and data sequences
For TCP, relate endpoints, flags such as SYN, sequence information and acknowledgements in time. Repeated SYNs without a visible response motivate hypotheses including an unavailable peer and an unseen return path. Repeated byte ranges can support a loss hypothesis, but one capture location does not identify the drop point by itself. Align the flow with application request timing and server observations. Normal TLS encrypts application content, so possessing packets does not reveal arbitrary HTTP processing detail or the server’s internal execution time. Distinguish what was visible on the transport path from what the application actually accepted, computed and committed.
Account for capture location and offload
Host captures can be affected by checksum and segmentation offload, producing a view different from final on-wire packets. An apparently invalid outbound checksum can be observed before the NIC fills it in. Capture drops, filter selection and snapshot-length truncation also affect evidence. Before concluding that a missing display proves no transmission occurred, check interface and namespace selection and capture-side loss. When comparing hosts, account for clock differences and match connection identity rather than correlating solely by timestamps. The capture mechanism has its own resource limits; an incomplete observation should not silently become a complete causal account of the network.
Preserve conclusions and unseen boundaries
Record what appeared at a specific location and time separately from the hypothesis explaining it. Seeing a connection request leave is different from proving the remote application received or processed it. If necessary, capture briefly at another boundary to narrow the missing-response interval. Capture files can contain sensitive addresses or plaintext credentials, so share only the necessary material. After a change, repeat representative requests at the same observation point and compare application success and delay alongside the packet behavior. Fewer retransmissions or a different packet count can support a conclusion, but neither alone establishes that the user-visible operation is now correct or timely.