On Wed, 07 Dec 2005, jonathan@dsg.stanford.edu wrote:
> In message <20051208014223.107493C015F@berkshire.machshav.com>,
> "Steven M. Bellovin" writes:
> [.. discussion of TCP reordering and vendors who got it wrong ...]
> 
> In fact, it's more than a feature, it's a _requirement_.
> 
> IEEE 802.3ad requires that all packets (frames, in layer2-speak) in a
> flow not be reordered (or htat they traverse the same path through a
> link-aggreation group, I forget the exact wording, but the intent is
> unmistakably clear. At least to anyone with our shared backgorund).

None of this, of course, means that you can't invent a new 'bonding'
protocol which encapsulates frames and guarantees in-order delivery
using the information in that encapsulation on the receiving end.

From TCP's point of view, 2*100Mb/s ethernets bonded in such a way
are 'just one big pipe'. 

Of course, for current hardware, Gigabit ethernet is probably cheaper to
obtain than two Fast Ethernet ports anyway. For speed, that's a better
choice.

If you need redundancy for the sake of redundancy, the fact that one
client<->server can't use all of both links probably doesn't matter. 

If you want speed too, in that case, just use 2*1000Mbs links, and you
won't care anymore, either.

-- 
David Maxwell, david@vex.net|david@maxwell.net -->
All this stuff in twice the space would only look half as bad!
					      - me