Packets Sent to the MultiNet Card

Packets sent to a switching card over network $i$ will be lost if the card is in a different network $j$ at that instant. In addition to sends, the buffering protocol has to ensure delivery of all packets sent to a MultiNet node. This requires nodes to buffer packets for MultiNet nodes that are currently unavailable, but which will become available in the near future.

Figure 2: Buffering in MultiNet

The MultiNet buffering scheme is illustrated in Figure 2. This scheme defines different behavior for APs and nodes. APs maintain the state of all the cards that are associated to it. Nodes instead maintain the state of directly reachable cards in any ad hoc network to which they are connected. The state table is a 4-tuple state space, containing information about the card's address, its Service Set Identifier (SSID) [10], the time when the card switched from this SSID, and when it is expected to switch back to this network.

The state table mentioned above is updated as follows. A card sends a packet to the AP in an infrastructure network, and to all nodes in an ad hoc network, just before switching from it. This packet informs all the reachable nodes in the current network of its temporary unavailability. The packet also contains information about the duration of absence from the current network and is used to maintain the state table at the nodes and APs. So, whenever a node or an AP has a packet to send to another node, it will first check the state table. If the destination card is currently associated to another network, the packet is buffered. When the state table indicates that the timer of a card has expired, and that it would have switched back to a network, the nodes and APs send the packets buffered for that destination on the corresponding network.

A concern for the buffering protocol is the loss of broadcast packets carrying state in the ad hoc network. Fortunately, as we will describe in Section 5.4, switching cards in ad hoc networks send more than one packet carrying this state per switching cycle. To handle situations when none of these messages get through, neighbors temporarily buffer packets for the switching node if it is unreachable. They estimate the activity period of the switching node based on past activity, and keep probing it. On receiving a response, they transmit the buffered packets.