Member
I wrote a Rust tool to detect path MTU blackholes. It sends padded UDP probes with DF set and times the retransmissions. Found a hosting provider in France doing 1280 on IPv4 "for consistency with IPv6." Consistently broken.
512MB RAM, 1280 MTU, what's the difference. Both are "plenty" until they aren't.
From a regulatory standpoint, interfering with ICMP for "security" while degrading basic connectivity is worth documenting. The BSI in Germany has guidance on this. Not that providers read it.
My office in Jakarta has a "business" connection from the incumbent. MTU 1460. Why 1460? "Standard for fiber." I asked about the 40 bytes. They said "overhead." What overhead? Silence.
1460 sounds like they subtracted TCP header from 1500 and called it the MTU. Classic layer confusion. I see this in Frankfurt colo sometimes, sales engineers who took one networking course.
I found VPS in Turkey with MTU 1200. Price was 2 USD/month. Support said "high security network." I said "you mean broken?" They said "same thing in Turkey." I stayed for the price.
I'm stealing that. New Jersey DMV runs on that philosophy.
Following. My home lab is all 1500 and I'm starting to feel like the weird one.
576 in 2019 was bad but at least it was a default. I just remembered: 2014, a WISP in rural Ontario running 512 byte MTU on their backhaul. Because "the radios prefer small frames." Their VoIP was... special.
Ubiquiti gear with default airMAX settings and someone who never changed the ethernet side. I've seen that exact pathology. The radio fragments everything anyway, so the small MTU just doubles the work.
