I agree, while wifi speeds get faster more and more bottlenecking will take place at the controller in a centralized architecture. What are some common ways to prevent this when CA is the given Architecture the customer wants. I know of etherchannel bonding.
Well these are always fun discussions. Do I like controllers? Yes and no. In my experience most of our traffic is either going to our datacenter servers or the internet, which is also at the datacenter. Phone traffic would be an exception when internal phone to phone traffic is in play. Internal to external traffic will still traverse some sort of demarcation at the core or datacenter. My understanding is that distributed architectures will still have a situations that require tunneling a client STA's traffic from one AP back to another leveraging something like a CAPWAP tunnel. Controllers do this as well be they are tunneled centrally. Another consideration is that your guest traffic will likely be segmented from you corporate traffic and aggregated centrally. Obvioulsy my point is that the controller was designed as a work around for the some things like L3 roaming. But they also do things like proxy ARP. ARP on the WLAN is bad. Broadcast traffic on network cards in general is bad. I think that the distributed model will be standard someday but it is still early in this technology. I'm sure vendors are solving the new problems associated with distributed architectures but it seems like this would still be early. Many vendors have taken a hybrid approach by putting filtering at the AP and using policy to determine if traffic is forwarded to a controller or dropped off at the edge switch. Well that's my 2 cents. Distributed architectures seem to be the future but I'm not convinced the future has arrived there are many problems to solve still.
I work for Ruckus and I think we have it right. :)
If you don't feel like reading the long answer, here's the short one. Controllers are here to stay; centralized forwarding is right, distributed forwarding is right.
I like the concept that Aerohive has, distributed architecture. Aruba did it the way they did for a really good reason. So, which is best?
The answer is, it all depends on what you are trying to accomplish. For Ruckus, the controller will NEVER go away. Now keep in mind, having a controller does not mean that the data is going through the controller.
Why is Ruckus never getting rid of the controller? A high number of our customers NEED centralized forwarding. In many cases a distributed architecture functionally wouldn't work, especially for our customer base. Is it the best architecture? In these cases, yes. It has downsides (central failure, bottleneck) but if you plan for it, it's manageable (but typically costs more money).
Now, all of our remaining customers do distributed forwarding.
The key is diversity. Our customers can choose central or distributed forwarding by SSID with a check of a box. Flexibility is key. Of course, in my opinion. :)
P.S. Repeat of my already public opinion: Aerohive will (or already has) centralized a box as a controller. :)
My 1 cents, having worked with the controller and "controllerless" architectures. As GT Hill says both have merits. We use both. We are diverse.
I believe that Centralized Management of the WLAN is a Must /Given. How you slice up the data and control plane "depends" on what you are trying to accomplish. I like what Ruckus is pushing with newly released ZD5000 and improving the ability to tunnel more of the traffic for security of certain flows- they do have the controller right and configuration is a SNAP. I like what Aerohive does with Cooperative Control and how you can manage from the cloud or the local domain. I like what Cisco does with Mobility Groups and Location Awareness. I like what a lot of vendors offer. I like to, learn , listen and let others make the call on what they LOVE too!
At the end of the day, look at the end device, which is the client. The customer cares about getting connected and the content to the device. If we as Wireless Experts and Engineers can deliver a secure and highly resilient Quality of Experience (QoE) to the end users with a level of security that meets their required posture, we have done much. WLANs are constantly evolving . The more devices made with these chip-sets, the more requirements/demand for access to the WiFi filling stations .
What is most needed, is good WLAN design and Hotspot Access portals to help with the IP Over Everything Explosion. That said, I end on this note: There is always a bottleneck, but the neck of the bottle can be expanded or stretched to accommodate to serve the ultimate goal. WiFi that works and scales for the users.
Thanks for all the comments. Just like many have stated, it all depends on the current situation and the specific needs of that customer. No one way is perfect for all. This is why we gather information before we design :-).