On Thu, 6 Nov 2025 at 19:52, nanog--- via NANOG <nanog@lists.nanog.org> wrote:
So you use header compression on all your links, right? No sense reducing your 1Gbps main uplink to 0.98Gbps. The checksum (removed in v6) is already 5% of each IP packet header. Speaking of headers I take it you're using SLIP instead of Ethernet? And you avoid TLS like the plague? I hope you replaced your 15W LED bulbs with 14.7W bulbs as well - your finance department will thank you. This is asinine.
IPv6 is kind of riddled with these changes for sake of change, which just made things worse. - removing checksum (no way to know when LSR/L2 transit has broken memory, mangling frames) - mandating multicast for L2 disco (not actually implemented, linux, cisco, juniper... doesn't even join the /mandatory/ group by default, switches do not have scale to handle it under adverse condition). In reality good 'ol flooding is used, because cure was worse than the disease - forcing ipsec into it, later removed - the whole lunacy of massive network sizes, making it impossible to make secure L2 disco without complicated and expensive protection logic which rarely exists. All while technically SLAAC and DAD would allow using arbitrarily small, or if you actually force L2 address in L3 address you could do stateless (no ND), which is also not implemented anywhere - whole next-header woe with TCP/UDP at the bottom, lucily in practice situation remains same as in IPV4, you cannot use those options and no new functionality can be added by introducing new ones, outside of controlled small environments, which might just as well run IPv42, when interop is not relevant. While in reality only utility needed or gained is more addresses, which is sufficient to fix the actual problem we had. But we keep trying to add opinionated solutions for problems that may or may not exist, adding disagreements and reducing motivation to implement and migrate. -- ++ytti