Phillip Hallam-Baker wrote:
I would hope that a practical benefit of IPv6 would be improved performance
as a result of support for large packets.
That is, like almost all the other ambitiously extended
functionality of IPv6, an illusion.
I had really hoped that IEEE would have made support for jumbo frames an
absolute requirement for all gigabit ethernet. But nooooo, its an option so
I'm not sure what is "an option", as maxUntaggedFrameSize of
Ethernet is 1518B without any option, though larger frames are
not actively prohibited.
I'm not sure what "jumbo frames" mean, as 9KB frames which is
available also to IPv4.
But, if you are talking about jumbograms, which is larger than
64KB, which is IPv6 specific, they are useless.
While some computers (vector super computers, especially) are
slow to react on interrupts and context switching to receive
packets, which was the motivation to introduce jumbograms,
such computers have I/O subsystem, which can take care of TCP
without bothering main CPUs.
Masataka Ohta
_______________________________________________
Ietf mailing list
Ietf(_at_)ietf(_dot_)org
https://www.ietf.org/mailman/listinfo/ietf