LAN devices intermittently lose connectivity after enabling jumbo frames on mixed‑vendor switches

Environment

Small office network with two stackable switches (VendorA model S24 v2, VendorB model X48), several Linux servers (Ubuntu 22.04) and ESXi hosts connected via multi‑gig LACP trunks; MTU changed from 1500 to 9000 on VendorA and ESXi host on 2026-09-12

Symptoms

Workstations and servers intermittently drop TCP connections to internal hosts (SSH, SMB) with rounds of 10–60 seconds of packet loss; large file copies fail or hang; repeatable after ~5–30 minutes under moderate traffic

Observations

Ping to gateway shows 0%→30% packet loss during incidents; ethtool -k on ESXi vmnic reports offloads enabled and MTU 9000; VendorB switch ports remained at MTU 1500; switch logs show occasional fragments dropped and 'LACP partner mismatch' messages; tcpdump captures on ESXi uplink show fragmented packets with DF cleared and retransmits; Windows clients show high retransmit counters

Already attempted

Rolled back jumbo frames on a non-production Linux server (ip link set dev eth0 mtu 1500) — connectivity stabilized for that host; cycled LACP on ESXi uplink and reconfigured static LAG — temporary improvement; updated VendorA switch firmware to latest stable — no change; rebooted VendorB switch during maintenance window — incidents persisted

Question

Why are some devices losing connectivity after enabling jumbo frames, and what steps isolate and resolve fragmentation/mismatch between mixed‑vendor switches without disrupting production traffic?

Post timeline

COMMUNITY RESPONSES

Verified community troubleshooting

0 replies

No answers yet..

Showing 0 of 0 community responses

AI DIAGNOSTICS

Incident analysis

Updated 1w ago• Trouble ID

Most likely causes

MTU mismatch across switches and ESXi physical NICs causing dropped jumbo frames 95%

NIC offloads (GSO/GRO/TSO) on ESXi creating large frames that intermediate switches cannot handle 75%

LACP/port-channel negotiation differences between switch vendors leading to inconsistent hashing or path MTU handling 70%

Summary

Symptoms indicate MTU mismatch / inconsistent jumbo frame support across the path (mixed vendor switches + LACP). When one hop uses MTU 9000 and another remains 1500, large frames are fragmented or dropped; LACP partner messages may indicate negotiation failures for aggregation settings. Fragmentation, DF handling, and offload interactions on ESXi can cause retransmits and partial outages for flows that exceed 1500 bytes.

Suggested Fixes

Diagnostic step 1

Document current MTU on every hop: on ESXi use esxcli network nic list and ethtool-equivalent output; on Linux servers use ip link show; on switches use show interface mtu or vendor CLI. Revert all devices to a consistent MTU (1500) as a safe baseline and verify stability before further changes.

Diagnostic step 2

In a maintenance window, enable jumbo frames end-to-end: set MTU=9000 on both switches, ESXi physical NICs, VM kernel adapters, and servers. Do one LAG/trunk at a time and monitor for drops.

Diagnostic step 3

Temporarily disable large send/receive offloads on ESXi and affected VMs (no data loss): on Linux VMs ip link set dev eth0 tx off rx off; on ESXi, consider advanced NIC offload settings. Check if packet fragmentation behavior changes.