This is a subject for another blog post. Or a book!
This problem has been excessively discussed in early IP days. The alternative to current model is to have a fixed packet size. This inherently means that for some media the packet size will be not optimal.
The current model in IPv4 is reasonable - that is: dynamic packet size, with either routers doing fragmentation or signaling mechanism to push back the PMTU onto the originating host.
The problem is that there is very little visibility if these mechanisms work. The linked online tests are an attempt to fix this - by easily showing everyone if their CPE / ISP / Path is reasonable and behaves correctly.
Testing fragmentation / ICMP PMTUD delivery historically was pretty hard. We need more visibility.
This problem has been excessively discussed in early IP days. The alternative to current model is to have a fixed packet size. This inherently means that for some media the packet size will be not optimal.
The current model in IPv4 is reasonable - that is: dynamic packet size, with either routers doing fragmentation or signaling mechanism to push back the PMTU onto the originating host.
The problem is that there is very little visibility if these mechanisms work. The linked online tests are an attempt to fix this - by easily showing everyone if their CPE / ISP / Path is reasonable and behaves correctly.
Testing fragmentation / ICMP PMTUD delivery historically was pretty hard. We need more visibility.