Choosing the right design for your network is a key to sucessfully deliver content to end users. Doesn’t matter if you’re just planing new network or you already have one, you probalby at least once doubt if your network design is right for your needs. Delaying the plan right design does not make problem smaller. 
This is a very simple topology just to be able to deliver content to end users. Ofcourse, this is not enough for ISP to function properly, as we need bandwidth shaping, authenticating and traffic control, dynamic routing protocol, firewall and many others. 
This is a more of a working network for ISP. Clients get authenticated on their AP using radius server that is connected on core router (logically or physically). Between 2 AP’s and rest of the network there is a RIP/OSPF routing protocol, which one exactly depends on size of the netowork and used equipment. Note that you cannot connect bigger network this way becouse you’ll need more routers to pass the content to APs. So, we need bridges that connect core to rest of the network. 
As the network grows bigger it’s harder and harder to keep it administrated and under control. So, you can break the network into logical segments responsible for it’s own role. This means – keep your AP config simple, to be able to simply add new AP and connect it to rest of the network, so you won’t need to enter 100’s of APs to change something. In the graph you can see that client traffic is ending on user switch instead of AP, so when you need to change something or perhaps introduce new way of authenticating users you can do it on just one place instead of chaning rules on many others. There is a few pppoe/hotspot concentrators that will be quite busy doing client authentication, firewalling and shaping. If you’re wondering why 3 concentrators and not just one strong enough for whole network, the answer is simple – you would be in quite a problem if that one is failing. In this network other routers will take the load of failed one on itself. Besides failovering, there is a load balancing. In PPPOE case (which is preffered over hotspot), pppoe clients will choose the first mac of pppoe concentrator that answers, and if you’re having three pppoe concentrators they have 33% chance to be elected for authenticating and creating pppoe sessions. You’ll need a routing protocol between pppoe concentrators and core router. You, however, don’t need to exchange local routes with network behind the switch as the connections are terminated on that switch. One thing to keep your mind on is the quality and features of that switch. All network traffic is going trough it, and that could means a tousands of learned mac addresses. It’s quite possible to overload it’s mac address memory buffer and downgrade it’s performance, so take a look at specs before chosing the right switch. Many people will recommend mpls on links over eoip and the reason is that mpls is more intelligent way to control your traffic but eoip is simpler and easier to implement. There is also an argument that mpls is using less cpu which i cannot agree in either real world or judging by complexity. On the other hand there is some strange behaivor with many EOIP tunnels that potentialy introduce unexplained lags. So choose what you think it suits you better.
As the right pppoe concentrator for you – choose the one you know to work with best. If that’s Cisco, go for it, if that’s plain Linux, go for it. As for the performance, it seems that modern x86 machines have alot power in cpu processing for this that which is not just a packet forwarding, so i would rather choose Linux-based solution.
Enterprise Networks 