Hi Michelot, I've just read your reply!
You can put a router in the 'core' of your network (the router, if I'm not wrong, realizes the Y junction, isn't it?) and everything will work! But this could not be the best solution:
here we face with your particular project; the number and type of network interfaces available on each devices, the 'type' and bandwidth of WAN links, the performance required among each node pair and other variables.
All these variables could lead to different (and sometime opposite) solutions: for example, if WAN links are leased lines you can manage them unconditionally; in this case you could put the ATM switch in the core, connect all the leased lines to the ATM to one side and to a router on the other and create the number and type of PVC (VP/VC)required (you can do this because YOU manage the ATM network). On the other side if WAN links are provided by an ISP they probably 'come' with a fixed and unchangeable PVC configuration: in this case a Router in the core of your network is mandatory and will be the only way to provide the flexibility and scalability you need!
You can use ATM in differnet ways: 'only' as a framing protocol OR as a framing + switching 'suite'; in your project you use ATM as a framing protocol over physical links: switching capabilities are fully damanded to layer 3 (that is, IP in the core router, that will decide how to route packets on a layer 3 and/or layer 4 information basis). Pros: extreme flexibility typical of IP. Cons: to be evaluated.
I usually prefer to design my project using ATM for its switching capabilities (like Ethernet) putting it everywhere is possible, beginning (mandatory) in the core's lower layer, and bringing IP to the edge, 'near the users'. This approach leads to a very performing and reliable network but, often, more expansive and complex to manage.
Hope this helps.
Bye
