Two MPLS service providers (SP1 and SP2) are offering inter-provider RFC 2547/4364-based IP- VPN service to an enterprise customer. IP traffic among some of the customer's sites has to traverse both o

Enabling PMTUD only on SP devices does not help because the enterprise senders still do not know to reduce their packet sizes before the traffic enters the MPLS network.

Enabling PMTUD on enterprise devices ensures that the senders set the DF bit and are equipped to process ICMP Fragmentation Needed responses, allowing them to reduce segment sizes when the MPLS label stack overhead causes packets to exceed an intermediate link's MTU. Because the enterprise senders are the origin of traffic, only they can reduce packet sizes at the source - making PMTUD on the enterprise the effective end-to-end solution.

Setting the DF flag alone without enabling the full PMTUD stack means senders will not process ICMP Fragmentation Needed replies or adjust their packet sizes, so large packets will still be dropped.

Clearing the DF flag permits fragmentation at intermediate hops rather than preventing oversized packets, and fragmentation in MPLS environments creates reassembly overhead and does not reliably solve MTU mismatches.