1
Diagnostic step 1
Set MTU to 1500 on server and both switches (vendor A and vendor B) and verify SMB reliability to confirm MTU is the factor (reversible).
FULL AI ANALYSIS
SMB shares become unreachable after enabling jumbo frames on mixed‑vendor switch
Generated Oct 12, 2026, 10:57:58 PM
Analysis summary
Symptoms point to path MTU/fragmentation issues crossing a segment that does not consistently support jumbo frames (vendor B switch). SMB (particularly older clients or NetBIOS components) is sensitive to dropped or malformed large frames. The fact that reducing MTU on one client immediately restored connectivity indicates the problem is frame-size related, and switch counters showing CRC/oversized frames implicate the intermediate vendor B device or an Ethernet trunk misconfiguration.
Possible causes
Partial jumbo frame support: one or more switches in the path do not support MTU 9000 or have inconsistent configuration causing frame drops.
95%Misconfigured trunk/aggregation: an 802.1Q trunk or LACP link between vendor switches may have MTU or fragmentation offload mismatches.
80%NIC offload mismatch: server NIC using large segment offload (TSO/GSO) producing jumbo packets that an intermediate switch mishandles.
65%Cabling or duplex issues causing corrupted large frames under high MTU settings.
40%Suggested fixes
2
Diagnostic step 2
Collect packet captures on server and on an affected Windows client while reproducing the hang; look for retransmissions, ICMP fragmentation-needed, and mismatched frame sizes.
3
Diagnostic step 3
Check and align NIC offload settings: temporarily disable TSO/GSO on the server (ethtool -K <iface> tso off gso off) and retest SMB to see if large segments cause the issue.
4
Diagnostic step 4
Examine switch port configurations and trunks: verify MTU and 802.1Q settings on the inter-switch links and ensure firmware on vendor B supports jumbo frames at configured MTU.
5
Diagnostic step 5
Inspect physical cabling on trunk/link aggregation for errors and replace cables on the path showing CRC errors before proceeding with further changes to avoid data corruption hidden by retransmits.