How to defeat the laws of physics
In my blog posting on enterprise WLAN, I talked about how the 2.4 spectrum simply does not have enough capacity for large scale deployments. My thesis was that users should really be looking to maximize the 5Ghz spectrum for their use (while recognizing that legacy devices will continue to exist in the 2.4 spectrum). Following this article, some vendors argued that existing WLAN architectures could not meet the needs of the enterprise, and the only valid architecture was one that used the same channel across all APs. They claim that a single channel architecture is the only one that can support fast roaming and provides maximum performance. In this blog series, I will explain why this is myth.Vendors of these single channel architectures claim that handoffs are too slow and cannot support the needs of real-time applications. I think it is important that we be intellectually honest when we discuss roaming speeds. First, I admit that in the old world of stand-alone APs, where no centralized coordination existed to help improve roaming times, this was an issue. There are two main functions that a client needs to do when it decides it will roam. First, it needs to find an alternative access point and then needs to provide its credentials to gain access to the network.Finding a candidate access point typically requires a client to scan all of the available channels, and on a dual mode (2.4Ghz and 5Ghz device), this is quite a number of channels. This process can be sped up through some innovation on the client side, but is best when it is assisted by the infrastructure. Cisco supports the neighbor discovery extension, which allows an access point to transmit its neighboring APs, along with their channels. This information can be used by clients to minimize the number of channels to scan.Once the client has found a new access point, it needs to request access, which includes the authentication process. The issue with re-authentication is the fact that it requires that the backend RADIUS server be involved, which increases latency, and the overall cost of the handoff. Eliminating the authentication process altogether, and only relying on a series of transactions between the client and the access point, is ideal. To do this, Cisco uses CCKM to allow a device to simply re-key, using information that was distributed during the original authentication phase. CCKM is supported on both the autonomous and unified wireless architectures, and the latter can support roaming times of sub-20ms.Cisco is working with the IEEE 802.11 to create an industry standards. Neighbor discovery will be available in 802.11k, while 802.11r will include CCKM- like features. This standard involves many of the consumer and enterprise class access point vendors, all of which are working diligently to provide these technologies to their customers quickly. Given that we’ve already proven that handoffs times can support the needs of real-time applications, I can only assume that the argument being made by these vendors is against access point technologies as they existed five years ago.Next time, we will focus on RF related issues brought up by these vendors.