by T. Sridhar, Flextronics
Deployment of Wireless LAN (WLAN) switches is increasing in enterprise networks. These devices, which can be standalone switches or integrated into a blade on an enterprise class switch, are useful for the management and control of WLAN access points. Although their deployment is a relatively new phenomenon, such control and configuration functions have existed before in WLAN controller devices.
WLAN switches connect to the WLAN access points (APs) through wired connections (through a switch port). They also connect to the enterprise network through their other switch ports. The switches are the “gateway” to the wired enterprise—all frames from WLAN clients have to pass through the WLAN switches to the enterprise network.
To understand the motivation for WLAN switches and their operation in the network, it is useful to view the WLAN network architecture and the functions of the access points. We can view the WLAN switch as the control function and the APs as the wireless termination function.
This article presents the function of WLAN switches and controllers by detailing WLAN network architectures along with functions of the AP and controller. It also presents the various functions on the controller to AP interface. Subsequently, it outlines variables related to Layer 2/3 mobility in the centralized architecture and concludes by presenting some common myths and reality about these architectures.
This article uses the term Wireless Termination Point (WTP) to refer generically to APs and the term Access Controller (AC) to refer generically to the WLAN control function (whether implemented on a WLAN switch or standalone controller).
WLAN Network Architectures
Three types of WLAN network architectures are commonly deployed:
The following sections describe these architectures in greater detail.
In the autonomous architecture, the WTPs completely implement and terminate the 802.11 function so that frames on the wired LAN are 802.3 frames. Each WTP can be independently managed as a separate network entity on the network. The access point in such a network is often called a “Fat AP” (see Figure 1).
During the initial stages of WLAN deployment, most APs were autonomous APs, and manageable as independent entities in the network. During the past few years, centralized architectures (discussed next) with ACs and WTPs have gained popularity. The primary advantage of the centralized architecture is that it provides network administrators with a structured and hierarchical mode of control for multiple WTPs in the enterprise.
The centralized architecture is a hierarchical architecture that involves a WLAN controller that is responsible for configuration, control, and management of several WTPs. The WLAN controller is also known as the Access Controller (AC). The 802.11 function is split between the WTP and the AC. Because the WTPs in this model have a reduced function as compared to the autonomous architecture, they are also known as “Thin APs.” Some of the functions on the APs are variable, as discussed in the following section (see Figure 2).
In the distributed architecture, the various WTPs can form distributed networks with other WTPs through wired or wireless connections. A mesh network of WTPs is one example of such an architecture. The WTPs in the mesh can be linked with 802.11 links or wired 802.3 links. This architecture is often used in municipal networks and other deployments where an “outdoor” component is involved. This article does not address the distributed architecture.
WTP Functions – Fat, Thin, and Fit APs
To understand the autonomous and centralized architecture, it is useful to look at the functions performed by the APs. We start with the Fat APs, which form the core of the autonomous architecture, followed by the Thin APs, which were specified as part of the WLAN switch- or controller-based centralized architecture. The article will then outline the functions of a new variant called the “Fit AP,” an optimized version of the AP for centralized architectures.
Fat Access Points
Figure 1 shows an example of an autonomous network with a fat access point. The AP is an addressable node in the network with its own IP address on its interfaces. It can forward traffic between the wired and wireless interfaces. It can also have more than one wired interface and can forward traffic between the wired interfaces—similar to a Layer 2 or Layer 3 switch. Connectivity to the wired enterprise can be through a Layer 2 or Layer 3 network.
It is important to understand that there is no “backhauling” of traffic from the Fat AP to another device through tunnels. This aspect is important and is addressed when discussing the other AP types. In addition, Fat APs can provide “router-like” functions such as the Dynamic Host Configuration Protocol (DHCP) server capabilities.
Management of the AP is done through a protocol such as the Simple Network Management Protocol (SNMP) or the Hypertext Transfer Protocol (HTTP) for Web-based management and a Command-Line Interface (CLI). To manage multiple APs, the network manager has to connect to each AP through one of these management schemes. Each AP shows up on the network map as a separate node. Any aggregation of the nodes for management and control has to be done at the Network Management System (NMS) level, which involves development of an NMS application.
Fat APs also have enhanced capabilities such as Access Control Lists (ACLs), which permit filtering of traffic for specific WLAN clients. Another significant capability of these devices is configuration and enforcement of Quality of Service (QoS)-related functions. For example, traffic from specific mobile stations might need to have a higher priority than others. Or, you might need to insert and enforce IEEE 802.1p priority or Differentiated Services Code Point (DSCP) for traffic from mobile stations. In summary, these APs act like a switch or router in that they provide many of the functions of such devices.
The downside of such APs is complexity. Fat APs tend to be built on powerful hardware and require complex software. These devices are expensive to install and maintain because of the complexity. Nevertheless, the devices have uses in smaller network installations.
Some Fat AP installations still use a controller at the back end for control and management functions. These controllers lead to a slightly scaled-down version of the Fat AP, called, not surprisingly, a Fit AP, discussed later.
Thin Access Points
As their name indicates, Thin APs are intended to reduce the complexity of APs. An important motivation for this reduction is the location of APs. In several enterprises, APs are plenum-mounted (and thus in hard-to-reach areas) so that they can provide optimum radio connectivity for end stations. In environments like warehouses, this is even more evident. For such reasons, network managers prefer to install APs just once and not have to perform complex maintenance on them.
Thin APs are often known as “intelligent antennas,” in that their primary function is to receive and transmit wireless traffic. They backhaul the wireless frames to a controller where the frames are processed before being switched to the wired LAN (see Figure 2).
The APs use a (typically secure) tunnel to backhaul the wireless traffic to the controller. In their most basic form, Thin APs do not even perform WLAN encryption such as Wired Equivalence Privacy (WEP) or WiFi Protected Access (WPA/WPA2). This encryption is done at the controller—the APs just transmit or receive the encrypted wireless frames, thereby keeping the APs simple and avoiding the necessity to upgrade their hardware or software.
The introduction of WPA2 necessitated encryption on the controller. Although WPA was hardware-compatible with WEP and required only a firmware upgrade, WPA2 was not backward-compatible. Instead of replacing APs across the enterprise, network managers could just backhaul the wireless traffic to the controller where the WPA2 decryption was done, and the frames were sent on the wired LAN.
The protocol between the AP and the controller for carrying the control and data traffic was proprietary. Also, there is no capability to manage the AP as a single entity on the Layer 2/3 network—it can be managed only through the controller, to which the NMS can communicate through HTTP, SNMP, or CLI/Telnet. A controller can manage and control multiple APs, implying that the controller should be based on powerful hardware and often be able to perform switching and routing functions. Another important requirement is that the connectivity and tunnel between the AP and the AC should ensure low delay for packets between those two entities.
With Thin APs, QoS enforcement and ACL-based filtering are handled at the controller—not a problem because all the frames from the AP have to pass through the controller anyway. Centralized control functions for ACLs and QoS are not new—they were implemented in networks with Fat APs too. Such installations have controllers that act as the gateway for managing traffic from APs to the wired network. However, the controller function takes on a new dimension with Thin APs, especially with respect to the data plane and forwarding functions. The controller function subsequently was integrated into Ethernet switches that connected the wireless and wired LANs—the motivation for the family of devices known as WLAN switches.
The Wireless MAC architecture in this scenario is known as the Remote MAC architecture. The entire set of 802.11 MAC functions is offloaded to the WLAN controller, including the delay-sensitive MAC functions.
Fit Access Points
Fit APs are gaining in popularity in that they try to take advantage of the best of both worlds—that is, the Fat APs and the Thin APs. A Fit AP provides the wireless encryption while using the AC for the actual key exchange. This approach is used for newer APs that use the latest wireless chipsets supporting WPA2. The management and policy functions reside on the controller that connects to multiple APs through tunnels.
Also, Fit APs provide additional functions such as DHCP relay for the station to obtain an IP address through DHCP. In addition, Fit APs can perform functions such as VLAN tagging based on the Service Set Identifier (SSID) that the client uses to associate with the AP (when the AP supports multiple SSIDs).
Two types of MAC implementations are possible with Fit APs, known as the Local MAC and the Split MAC architectures. Local MAC is where all the wireless MAC functions are performed at the AP. The complete 802.11 MAC functions, including management and control frame processing, are resident on the APs. These functions include time-sensitive functions (also known as Real Time MAC functions).
The Split MAC architecture divides the implementation of the MAC functions between the AP and the controller. The real-time MAC functions include functions such as beacon generation, probe transmission and response, control frame processing (for example Request to Send and Clear to Send—RTS and CTS), retransmission, and so on. The non-real time functions include authentication and deauthentication; association and reassociation; bridging between Ethernet and Wireless LAN; fragmentation; and so on.
Vendors differ in the type of functions that are split between the AP and the controller, and in some cases, even about what constitutes real time. One common implementation of a Fit AP involves local MAC at the AP and control and management functions at the AP.
Access Controller and Control Functions
The next critical component of the Centralized WLAN Architecture is the Access Controller (AC). For the following discussion, we consider the controller function to be implemented on a WLAN switch and call the function an AC. We also use the term “WTP” to refer to APs (fat, thin, or fit).
The Control and Provisioning of Wireless Access Points (CAPWAP) Working Group in the IETF is working on defining the interface and protocol between an AP and its controlled WTP. This section uses the CAPWAP framework to detail the interface between the AC and the WTP. [3,4,5]
Figure 3 shows an enterprise network with multiple ACs and WTPs. The WTPs can be connected to the ACs through a Layer 2 (switched) or Layer 3 (routed) network. The interface between the WTP and the AC is responsible for the following:
In addition, although CAPWAP does not explicitly define all the details, the AC performs functions such as Radio Resource Management (RRM) and rogue AP detection based on configuration and monitoring of the various access points in its domain of control. The extent of these functions varies according to the vendor implementation. Another important function provided by ACs is mobility management. The following sections provide more detail about these functions, with specific reference to CAPWAP. Note that the CAPWAP protocol, which is based on the Cisco Lightweight Access Point Protocol (LWAPP), is still under development in the IETF, as of the writing this article (March 2006).
Discovery and AC Selection
A WTP discovers an AC to connect to through discovery request messages, to which one or more ACs can respond (depending on the network topology). Communication between the AC and the WTP is through the User Datagram Protocol (UDP). The WTP determines which AC to connect to and then tries to establish a secure session with the AC. Subsequent CAWAP packets are sent over the secure session.
Subsequently a configuration exchange takes place between the AC and WTP. This exchange includes:
CAPWAP control messages include the following message types:
Configuration includes information such as beacon period, maximum transmit power level, Orthogonal Frequency Division Multiplexing (OFDM) control, antenna control, supported rates, QoS, encryption, and so on.
AC and WTP Interaction
The WTP provides information such as hardware, software, or boot version; maximum number of radios; radios in use; encryption capabilities; type of radio (802.11b/g/a/n); type of MAC (local, split, or both); tunneling modes; and frame type between AC and WTP (for example, local bridging or native bridging—that is, encapsulating all user payloads as native wireless frames).
The AC information includes hardware or software version, number of mobile stations currently associated with the AC, number of WTPs currently attached to the AC, maximum numbers for each of these, security parameters (authentication credentials) between AC and WTP, control IPv4 or IPv6 address, and so on.
Because the WTPs fall under the category of “Fit APs,” they can also be configured with an IP address from the AC. Another parameter that can be configured is ACLs at the MAC address level.
Rebooting (reset) of the WTP can be done by the AC at any time. Independently, the WTP can request a new image through an Image Data Request, which is followed by an Image Data Response and the image data itself.
Events are sent by the WTP when it determines that it has important information to send to the AC. Such information can include data transfer messages that can be used to deliver debug information from the WTP to the AC.
Radio Resource Management
Radio resource management is a generic term used to describe the control and configuration of radios on the AP. The type of control includes reducing and increasing the strength automatically or on user input—for example, if two WTPs controlled by an AC are interfering with each other, the AC can send a signal to one of the APs to reduce its strength. It can also do this based on user configuration.
Several WTPs are designed to also be used as “Air Monitors;” that is, they can monitor channels when not transmitting. Opinion is still divided on whether this mode of using WTPs is efficient—some vendors use dedicated air monitors instead of having their WTPs do double duty. With dedicated air monitors, it is much easier to scan and monitor all channels without having to worry about degrading the service for client stations.
Air monitors can forward information about other access points to the AC. The AC can determine if the information is for a valid WTP (that is, one that is supposed to be on the network and has, in fact, registered with the AC) or for a “rogue” access point. If it is for a rogue access point, the AC can perform multiple steps to prevent clients from attaching to this AP—for example, it can instruct the air monitor to “jam” this rogue AP by increasing the transmit power on the same channel.
Mobility management can take two forms—Layer 2 and Layer 3 mobility. Consider a client moving from one WTP to another, a scenario that can happen when a user with a laptop moves between two conference rooms within the same building. The client station reassociates with the new WTP, after which authentication is performed. Note that the association with the previous AP is “broken” before the association with the new AP is “made;” thus handoff in WLANs is known as “break before make.” Although this approach can lead to potential traffic disruption (and retransmissions), it is chosen over “make before break” (used in cellular telecommunications) to keep the client radio simple and less expensive.
One way to envision Layer 2 and Layer 3 mobility is to treat Layer 2 mobility as movement between APs under the control of the same AC (that is, Layer 3 network), whereas Layer 3 mobility is movement between APs under the control of different ACs.
Layer 2 Mobility
Layer 2 mobility means that when the station moves from one WTP to another, there is no impact on the IP addressability, effectively meaning that all the APs are on the same Layer 2 network and implying that they are connected to the same AC (see Figure 4). To prevent loss of data destined to the Layer 2 client, the WLAN switch mustnow forward client data to the new WTP. After the client association, the new WTP sends out an Ethernet frame to the AC with the client’s MAC address as the source address. The switch now associates the client’s MAC address to the port on which the new WTP is connected.
Although this process works well with Layer 2 (switched network) connectivity between the APs and the AC, it requires a slightly different approach when tunnels are used between them. The AC moves the mapping of the client to a different tunnel (that is, a virtual port) when it receives the MAC frame from the new WTP.
Another concern to be addressed with Layer 2 handoff is the buffering of data at the WTP. In normal circumstances, the switch or AC is not aware of the handoff until it hears from the new WTP. However, with enhanced statistics available at the WTP, it can determine that the specific client has moved away from the old WTP and stop forwarding data to the old WTP. These statistics can include maximum retry attempts on the Carrier Sense Multiple Access/Collision Avoidance (CSMA/CA) MAC layer protocol on the wireless link. The switch does not need to buffer the data because it is not clear when the handoff to the new WTP will occur. This approach helps avoid wasteful traffic on the link between the old WTP and the AC.
Some vendors have approached this problem differently with Fat APs. There, the APs might buffer the traffic until they see a frame from the switch indicating that the client is now on a different switch port. These APs then send the buffered traffic to the switch, which forwards that to the new WTP. Because our intent is to lower the complexity of the WTPs, this approach is not a preferred one in the Centralized AC + WTP architecture.
Another important feature of Layer 2 roaming is preauthentication that needs to be done on the new WTP. Through 802.11i, clients can preauthenticate with neighboring WTPs so that roaming to a different WTP does not involve the lengthy authentication process of Pairwise Master Keys (PMKs) being sent to the new WTP. (The Pairwise Transient Keys (PTKs) still need to be derived.)
When the AC maintains the PMK for a specific client (through interaction with a RADIUS server), this process is automatic—that is, the AC can send the client-specific PMK to the new WTP. The encryption of 802.11 frames is still done by the old and new WTPs with the new PTKs.
Layer 3 Mobility
Layer 3 mobility involves the client retaining the same IP address while moving across multiple APs. This often happens when the client has published its IP address to multiple nodes. Such a scenario is likely in peer-to-peer communications and when the mobile station needs to act as a server for some function. It is desirable that the correspondent nodes communicating with the mobile node not have to change their configuration whenever the mobile node moves to a new Layer 3 network.
This problem of Layer 3 mobility is solved by Mobile IP . We do not discuss the details of Mobile IP here except to indicate that it has three distinct components. The Home Agent (HA) on the client’s home network is responsible for the address of the client. All packets destined to the client’s (invariant) IP address are sent to the Home Agent. If the client is on the home network, the HA forwards the packets directly to the client. If it is on a foreign or visited network, the HA forwards the packets to a Foreign Agent (FA) that is on the visited network.
To do this, it has to set up a tunnel to the FA—which is usually a Generic Routing Encapsulation (GRE) or IP-in-IP tunnel.
After stripping out the original packet from the tunnel, the FA is responsible for forwarding the packet to the client. This description is a simplification—numerous other steps are involved here. The important factor in a wireless LAN scenario for Layer 3 client mobility is where the Mobile IP endpoint resides. Some client stations include a software stack for a MIP client.
This Client MIP (CMIP) software:
The CMIP approach was the recommended approach for implementing MIP. However, it has the disadvantage of having to add a MIP client to every mobile station in the network—a setup that can become cumbersome when there are a large number of mobile stations.
The Centralized AC + WTP architecture offers a way of alleviating this problem. Some AC/WLAN switch vendors have implemented the MIP function on the AC so that the client never needs to be changed. Some implementations call this a Proxy MIP function.
The AC acts as an FA to terminate the tunnel from the HA and also performs the translation of the packets to the client’s address on the visited network when forwarding packets to the client. When the client sends Layer 3 packets out, it sends them through the AC, which, in turn, modifies the headers for the source IP address and tunnels the packets to the HA. This process is called “reverse tunneling” (see Figure 4).
When you consider a large enterprise network topology with multiple ACs and APs, you can envision the MIP tunnels to be established between the various ACs. (That is, they act as Foreign Agents for one set of users and as Home Agents for another set.) From a scalability perspective, it is important that the ACs have the necessary horsepower and switching capability (switching between tunnels from the APs to the ACs to the tunnels between the ACs).
WLAN Switches and Centralized Architectures – Common Myths
Previous sections considered various aspects of the Centralized AC + WTP architecture and some of the implementation factors. This section outlines some common myths about these architectures and implementations. The intent is to examine this still-evolving area to facilitate clarity.
This article has provided the functions and deployment of WLAN switches by detailing the architectures that rely on a centralized controller managing a set of wireless termination points. It outlined some major aspects of the CAPWAP control functions and the concerns related to Layer 2 and Layer 3 mobility while implementing an AC + WTP architecture. Although protocol standardization is being done in the IETF for this emerging area, there is still sufficient scope for vendor differentiation.