Moving Toward Flat Networks

Question

What exactly is meant by a flat network? Please read on before you give a quick answer.
I have some definite ideas, but I would love to hear other people's inputs, ideas and perceptions.
The   trend in the data center is to go in the direction of flat networks,  which to me means  that the multi-tiered architectural model can be  collapsed into 2  layers. Juniper says they will collapse the data  center into one tier:  the Stratus project promises the equivalent of  one gigantic logical  switch.
So, the implication is that there will be a flattening of the network in 2 ways: first, in terms of physical layers of networking and secondly, in terms of creating one logical L2   domain that is horizontally expanded (across data centers, too), but   the redesigned L2 domain will not require STP, blocked uplinks, or   unknown unicast flooding.
I do require some clarification:
1.)   How exactly can the access and aggregation layers be flattened? If,  for  example, one can stack all the top of rack (accesss layer) switches  to  create a single virtual chassis and then do the same, or something  similar, like VSS, to the aggregation  layer, the result is that each  layer will see the other layer as one  switch. So, 10 access switches  will look like 1 switch to the  aggregation layer, and 2 aggregation  layer switches will look like 1  switch to the access layer. This will  allow one to create multi-chassis  etherchannels that can expand the  network horizontally, HOWEVER, there  are still 2 layers of networking.
Does this make sense?
2.)   The value in spanning L2 domains across data centers in a  services-oriented architecture is clear. It facilitates vMotion and  vStorage and creates clusters of compute and  storage resources, which  can be leveraged for SAN replication, disaster  recovery, cloud  computing (IT/Software as a Service), and running active/active  application services).
But  does  flattening the network mean that there will be one massive subnet  that  will be considered one broadcast domain? Albeit, unicast flooding  and  ARPs will have to be re-engineered to operate in this environment.  This doesn't seem likely.
Any   thoughts?
Thanks

Answer 1

There are as many ways to skin the cat as you can dream up.
But clearly you cannot expect to change roles completely and have everything stay the same.. although the Mac should not need to start over.. TM should be able to have multiple setups where it recognises the different network layouts.. at least with ML it can do that. If you are using SL as per your profile.. no.. it has to be manually setup.
The first obvious solution is this.
Network B: Mac connects to LAN via a non-Apple wi-fi router; TC connects to Mac via ethernet cable
Do the same in network A.. Use the Mac by wireless for internet.. and backups to the TC connected directly by ethernet. Do not plug the TC into the network.
A variation on this.
You can even do it by ethernet.. simply have a switch and plug in both Mac and TC.. but use the TC on a different IP address. You can set statically double IP on the one ethernet port.
Even better plug in a USB drive and use that.. it will be faster, more stable, and more reliable for backups than the TC. Leave the TC in network A.