These asumptions frame the rest of this article:

  • Many applications recover gracefully from path MTU discovery (PMTUD) blackholes
    • A non-trivial portion of applications don’t
  • MTU dysfunction abounds in modern networks
  • There are valid use-cases for jumbo Ethernet
  • PMTUD is preferable to manually decreasing interface MTU
  • Maximizing operational stability and efficiency in delivery of IT sericees is desirable.
  • The cure should never be worse than the disease
  • Packet loss on IP networks leads to application performance problems, which adversely impacts operational stability
    • MTU mismatches are an easily remediated cause of undetected packet loss
  • Path MTU Discovery (PMTU) provides a mechanism for two hosts on separate subnetes/broadcast domains with an MTU mismatch between them to compensate and avoid adverse impact
    • But PMTUD “in real life” is a fragile framework
  • Ethernet and IP networks represents the overwhelming majority of enterprise networks
  • Commonly adopted networks design practices result in widespread MTU mismatch
  • Modern public/private cloud hosting architectures often involve multiple layers of nested encapsulation as part of recursive underlay/overlay services
  • Remediating MTU mismatch introduces its own risks to operational stability
  • Most devices on enterprise networks have explicitly configured network interface MTU values
  • Most devices on enterprise networks consume DHCP services for IP address assignment