Re: [FE] Frame size poll
Kevin,
Before everyone gets too excited...
I'm not sure whether there would be a universally acceptable number for
the "tag overhead." Developers and designers would be well advised to
look carefully at their favorite architectures and see what the "good"
number would be for them. I would also hope that the group as a whole
would take a tolerant attitude and would adopt a number that is
universally palatable rather than going for the number that can reach
75% at the expense of 25% misery.
Secondly, I was looking at this from an architectural point of view. An
architecture may be capable of supporting a frame size increase to 2000
bytes seamlessly but this does not imply that currently implemented
systems will be able to. I expect that a number of people would prefer
to see an increase that doesn't obsolete currently operating systems.
This would most likely lead to a much smaller number (1522 bytes seems
good :-). I haven't done any research into what number would give
reasonable coverage of existing systems or whether this requirement is
even reasonable.
That said, 2000 is a very round number.
Hugh.
Kevin Daines wrote:
>Hugh,
>
>You've hit on something I was hoping would surface soon. In your opinion, is there value in trying to determine this "memory tag" (hoping this newly coined term 'memory tag' won't get confused with any other tag) size on existing gear as a means of gauging impact to existing implementations?
>
>I am familiar with implementations that store full frames and have 1,536 octet upper limited limits. This allows 18 octets of memory tag space for max untagged frames or 14 octets of memory tag space for max VLAN tagged frames.
>
>Is a memory tag size of 48 octets a safe upper limit? If so, 2,000 could be the max frame size. 1500 for max data payload and 500 for everything else. Hmmm...sounds like the original Ethernet goal of 1K data payload and 500 odd octets for everything else...
>
>Determining the new max frame size is one question. Deciding if these larger frames will be "enabled" on some links and not others is a separate, but also intesting, question. Perhaps I'll save that for another poll...
>
>Kevin Daines
>Chair, 802.3 Frame Expansion Study Group
>
>
>
>
>