Owen DeLong wrote:
Given the combination of Moore's law and the deployment lifecycle, designs we do today in this regard can be expected to last ~12 years or more, so they should be prepared for at least 16x. At 1,600 Gbps, that puts our target maximum MTU up around 200M octets.
If ICMP is the only interop method we have, we will still be using 1500 with 1280 the next most popular value. L2 uniformity requirements is a contributing factor to this as well as to the various ways ICMP does not serve well for L3. Consider a NBMA networks, or commonly found today, NHRP tunnela. All endpoints must use the same MTU. Perhaps some devices are capabable of much higher. Perhaps some tunnel egresses are capable of re-assembly. Static configuration of MTU is not ideal. It does not allow for dynamic changes in circumstances, such as tunnels/encapsulation rerouting between different MTU links. Between mpls and ethernet services, encapsulation and re-encapsulation is more and more prevalent, making MTU an issue again and again. Blocking is not the only way ICMP messages may fail to arrive. Configured MTU values can very easily be incorrect and hard to detect - this will get worse as L2 networks are glued together through various providers and segments. Devices in the path may have rate limiting or may simply be unable to route back to the source. Devices should be able to dynamically detect MTU on L3 links, and yes, maybe even for L2 adjancencies as well. L3 protocols should not have to rely solely on ICMP exception handling to work properly. Joe