This thread relates closely to something I posted on li this morning:

Paul Vixie
https://www.linkedin.com/feed/update/activity:7478870998654844928
3h

In infotech, laziness is a primary virtue. However this sometimes creates more 
problems than it solves. I'm thinking today of the 1500 octet packet size of 
Ethernet.

In ten megabit over coax, 10base5, the preamble and other analog framing gave 
us 14400 pps for minimum sized minimum spaced packets on an unshared 
transmission medium and around 813 pps for maximum sized minimum spaced packets.

We have clock *= 10'd many times since then, and sometimes *= 40 or *= 2. No 
change was made to minimum or maximum packet size, since each new Ethernet 
speed had to bridge to previous whenever speeds. This means pps *= factor every 
time, and on 400 gbit the pps is so fast we need a real time os with it's own 
CPU to manage the connection. Folly!

We should have given sqrt(factor) to pps and sqrt(factor) to maximum packet 
size. If ip4 and ip6 had mtu negotiation both on the local network and the 
global network, everything would today be faster and cheaper including our 
electric power costs.

It's not too late because it's never too late especially on something like this 
which is known to be eventually inevitable. But it would take collective will 
and vision, and money has changed everything.

Anyway that's why RFC 9715 now exists.
_______________________________________________
DNSOP mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to