Wrong math. The internet average packet size is very close to 750B - it has 
been published many times in many places.

Wrong assumption about the user needs. User does not care about serialization 
time. He/she cares about FCT==Flow Completion Time.
If one would get 2.6% less on the bottleneck, then his FCT would be 2.6% longer.
His page would open later. His file would download later.

Ed/
> -----Original Message-----
> From: William Herrin <[email protected]>
> Sent: Tuesday, December 2, 2025 23:37
> To: North American Network Operators Group <[email protected]>
> Cc: Marco Moock <[email protected]>; Vasilenko Eduard
> <[email protected]>
> Subject: Re: IPv6 Performance (was Re: IPv4 Pricing)
> 
> On Mon, Dec 1, 2025 at 11:09 PM Vasilenko Eduard via NANOG
> <[email protected]> wrote:
> > Fundamentally, IPv6 should be slower because of the bigger
> headers/overhead.
> 
> Hi Vasilenko,
> 
> Everything else being equal, IPv6 would have roughly 1.3% less throughput.
> The math is straightforward: 20 bytes larger header on
> 1500 byte packets, 20/1500 = 0.013 for throughput.
> 
> The latency difference is determined by the total packet size including header
> and data which is the same or smaller with IPv6 -- both are limited by the 
> 1500
> byte Ethernet frame size. Even on the initial packets smaller than the frame
> size, the difference in transmission time over gigabit or better links is so 
> small
> it disappears into the noise. So, no impact at all.
> 
> That's it.
> 
> If you're experiencing slower IPv6 or slower IPv4, it's all about the network
> engineering. Your network path to a server which satisfies one address is
> longer or transits a slower link in the path than the one which satisfies the
> other.
> 
> Regards,
> Bill Herrin
> 
> 
> --
> For hire. https://bill.herrin.us/resume/
_______________________________________________
NANOG mailing list 
https://lists.nanog.org/archives/list/[email protected]/message/TAMQ2C47CSSC7MFSQJDL22VMRZMLOQHZ/

Reply via email to