← All insights
ENGINEERING JOURNAL / Network Operations

Small requests work but transfers stall: investigate path MTU

Diagnose size-dependent network stalls with path-MTU evidence, tunnel overhead and permitted ICMP feedback before changing every interface or masking the symptom.

Small requests work but transfers stall: investigate path MTUSmall requests work but transfers stall: investigate path MTU — conceptual 3D-style vector showing Small Packet, Tunnel Limit, Mtu Feedback, Large Transfer.. Conceptual teaching illustration; connections show the investigation flow rather than a vendor deployment topology.CLOUVENIX / NETWORK OPERATIONSSmall requests work but transfers stall: investigate path MTU01Small Packet02Tunnel Limit03Mtu Feedback04Large TransferConceptual investigation flow / Small requests work but transfers stall: investigate path MTU
Small requests work but transfers stall: investigate path MTU — conceptual 3D-style vector showing Small Packet, Tunnel Limit, Mtu Feedback, Large Transfer.

Reachability can depend on packet size

A successful small request does not prove that a path can carry larger packets in the same way. Encapsulation, a lower-MTU link and missing feedback can create size-dependent failures. An application may connect successfully and then stall when it transfers a certificate chain, file or larger response.

Path MTU is the usable packet-size limit across a particular route, not simply the number configured on the application's nearest interface. This guide investigates that path with the network owner. It does not prescribe a single replacement MTU for every VPN, IPv4 route or IPv6 deployment.

Setup: preserve the working and failing flows

Record the source, destination, IP version, ports, route and any tunnel or overlay. Compare a small and larger operation over the same intended path. Note whether only one direction fails, because return traffic can follow different links or encounter different filtering.

On a suitable Linux diagnostic host, a read-oriented starting tool is:

tracepath -n example.internal

Use the approved target rather than the illustrative name. Tool availability and the usefulness of its probes depend on the host and network policy. A missing response from an intermediate hop does not by itself establish a broken MTU path; correlate it with actual traffic and the relevant endpoint evidence.

Illustrative incident: a VPN introduces a smaller usable payload

In this fictional incident, a service works on the local network but larger responses stall through a new encrypted tunnel. Basic connectivity and short requests succeed. The network team examines tunnel overhead, permitted control messages and the actual packet-size behavior rather than immediately changing application timeouts.

The comparison uses the same service operation on the working and failing paths. It establishes a testable hypothesis about packet size, not a conclusion that every VPN stall is an MTU defect. Congestion, retransmission and application limits remain separate possibilities until evidence distinguishes them.

Troubleshooting: connect size, path and feedback

ObservationEvidenceQuestion
Small requests succeed and larger ones stallComparable flow tests and endpoint capturesIs failure correlated with packet size?
Issue appears only through a tunnelEncapsulation design and effective pathDoes overhead reduce the usable packet size?
Sender repeats without useful progressPacket and permitted control-message evidenceIs MTU feedback reaching the sender?
One direction behaves differentlyForward and return path comparisonAre the routing or filtering boundaries asymmetric?

IPv4 and IPv6 path-MTU discovery have different protocol details. Follow the correct procedure for the observed IP version instead of applying an IPv4-specific assumption to IPv6. Keep packet captures access-controlled and limited to the diagnostic scope.

Correct the evidenced network boundary

Review the tunnel or interface design, required ICMP feedback and any supported transport adjustment with the network and security owners. Permit the specific control traffic needed by the design rather than removing filtering broadly. Where a transport workaround is justified, document the traffic to which it applies and the limitation it addresses.

Avoid reducing MTU everywhere as an unexplained permanent fix. That can conceal the problematic path and affect unrelated workloads. Define a rollback, test both directions and retain a representative large-transfer check for the route that originally failed.

Verification, alerting and SOP

Verify small and large transfers, connection setup and the original application operation over the repaired path. Compare retransmission or stall evidence before and after under a similar workload. A successful ping alone is not acceptance of the original problem.

Monitor path-specific transfer failures and relevant tunnel changes, using synthetic tests where appropriate. An internal SLO can describe completion of representative transfers on supported paths; a contractual network SLA needs separate terms. The SOP should record IP version, routes, encapsulation, approved tools, packet-size evidence, security-owner review and the test that demonstrates recovery.

See Clouvenix network architecture for a reviewed implementation and operational handover.

Technical reference

iputils tracepath manual, IETF IPv4 path MTU discovery, IETF IPv6 path MTU discovery. Reviewed 19 October 2026; use documentation for the installed product and release.