Thursday, May 6, 2010

Layer 3 Roaming

Layer 3 mobility is a superset of Layer 2 mobility. An 802.11 client must perform a Layer 2 roam, including AP discovery, before it can begin a Layer 3 roam. This section focuses on issues surrounding Layer 3 roaming, specifically with the IP Protocol and Mobile IP extensions (RFC 2002). It covers the following topics:
  • Roaming between roaming domains
  • A Mobile IP overview

Roaming Between Roaming Domains

As previously discussed, a roaming domain is defined as APs that are in the same broadcast domain and configured with the same SSID. Stated another way, a client can only roam between APs in the same VLAN and with the same SSID. As WLAN deployments expand within an organization, roaming domains might need to scale beyond a single Layer 2 VLAN.

Consider the following scenario: Company A has a four-story building in which it has deployed a WLAN. The initial deployment was small, and the WLAN was a single Class C subnet for the entire building. This setup created a roaming domain across all four floors of the building. As time progressed, the number of users increased to the point that the subnet is full, and performance is degrading because of increased broadcast traffic.

Company A decides to follow its desktop subnet model and use a single subnet per floor for the WLAN. This setup introduces complications because now the roaming domains are restricted to a floor, not the entire building as before. With the new subnet model in place, application persistence when roaming across floors is lost. The application most impacted is Company A's wireless VoIP devices. As users move between the floors (and subnets) on their wireless phones, they drop their calls when they roam. Figure 5-8 illustrates this scenario. In this figure, an 802.11 VoIP phone is connected to a wired VoIP phone. As the user roams from AP1 on Subnet 10 to AP2 on Subnet 20, the session drops because the roaming user is now on a different subnet.


Mobile IP Overview

The scenario described for Company A is common. Many applications require persistent connections and drop their sessions as a result of inter-VLAN roaming. To provide session persistence, you need a mechanism to allow a stsation to maintain the same Layer 3 address while roaming throughout a multi-VLAN network. Mobile IP provides such a mechanism, and it is the standards-based, vendor-interoperable solution to Layer 3 roaming for WLANs.

A Mobile IP–enabled network has these key components:


Agent Discovery

A roaming MN must dtermine that it is on a foreign subnet in a timely manner to minimize delay to running applications. HAs and FAs advertise their services by using the Internet Control Message Protocol (ICMP) Router Discovery Protocol (collectively known as IRDP) messages to send agent advertisements. As the MN establishes connectivity to the subnet it roams to, it listens for the periodic IRDP packets. The packets are sent to either the all-host multicast address (224.0.0.1) or the limited broadcast address (255.255.255.255). The IRDP packets are not sent to the subnet-specific broadcast address because the MN might not be aware of the subnet it has roamed to. In addition to periodic agent advertisements, an MN can solicit for advertisements after it detects that its interface has changed.

The agent advertisement contains two fields that allow the MN to determine whether it has
roamed to a new subnet:
  • The lifetime field from the agent advertisement
  • The prefix-length extension

The lifetime field provides a time value that an agent advertisement is valid for. If no new advertisement has been received before the lifetime reaches zero, the MN should attempt to discover a new agent.

The prefix-length extension indicates the network address value of the advertising agent. A change in prefix length (indicating a change in network address or subnet) shows the MN it should attempt to discover a new agent.

Upon determining it is on a foreign subnet, the MN gleans the CoA from the agent advertisement. The CoA can take two forms:
  • The address of the FA.
  • CCoA (Note that the CCoA is not advertised by the FA, but it is probably acquired by the MN as a Dynamic Host Configuration Protocol [DHCP] option.)

A CoA pointing to the FA forces the FA (usually the subnet router) to handle Mobile IP administration for all foreign MNs on the subnet in addition to handling packet-forwarding duties. The benefit to this situation is that only a single tunnel is required from the HA to each unique FA.

A CoA that is temporarily assigned to the MN places the Mobile IP administrative burden on the MN and forces the HA to establish a unique tunnel to each roaming MN. Figure 5-11 contrasts these two methods.