Chapter 3: Packet Switching (overview) Dr. Rocky K. C. Chang 23 February 2004

1. The problem statement How do two hosts that are not directly connected communicate with each other? Some “intelligent” packet forwarding (or switching) engines have to be in place between the two hosts. Chapter 3 vs. Chapter 4: Chapter 3 handles the problem on connected networks of the same or very similar type, e.g., both hosts on a switched Ethernet LAN (or ATM). Chapter 4 handles the problem on connected heterogeneous networks, e.g., one host on an ATM network and the other on an Ethernet LAN.

1. The problem statement Host D Switch 1 Host E Host F 3 1 2 Switch 2 Switch 1 Host E Host F 3 1 2 Switch 2 Host C 2 3 1 Host A Host G Switch 3 Host B 1 3 2 Host H

Layer-two or layer-three switches 2. The problem statement Layer-two or layer-three switches Switching based on the layer-two or layer-three information (headers). The switches between a source and a destination “work together” to deliver packets between them. Approaches to switching packets: Packet switching (a pure connectionless approach), e.g., switched LANs and IP. Virtual circuit (or cell) switching (need connection setup), e.g. ATM.

3. Switching and forwarding A switch is a multi-input, multi-output device, which transfers packets from an input to one or more outputs. LAN (layer-two) switches, e.g., switched Ethernet, switched token-ring, switched FDDI, ATM switches, and routers (layer-three switches). In addition to switches, need network protocols for switches to make forwarding decisions.

Packet switching refers to a pure connectionless approach. Each packet (header) contains enough information for the routers to make forwarding decisions. When a packet is received, the switch attempts to match it with an entry in its forwarding table. Each entry in the forwarding table mainly consists of destination address, and the interface (or port) where this packet should be forwarded to.

3.1 Packet switching Host D Switch 1 Host E Host F 3 1 2 Switch 2 Switch 1 Host E Host F 3 1 2 Switch 2 Host C 2 3 1 Host A Host G Switch 3 Host B 1 3 2 Host H

3.1.1 Properties of packet switching Forwarding decision is made packet by packet, even when several packets belonging to the same “stream.” Each packet should contain sufficient information for forwarding purposes. No connection set-up is required before sending packets. The design for the source is very simple. The intelligence lies with the switches. Intermediate switches do not keep any “states” about a certain “connection.”

3.1.2 Routing problem in packet switching Routing problem: How do the switches construct their forwarding tables? For example, Ethernet LAN switches are required to “remember” the frames they saw and to run a spanning tree protocol among themselves. IP routers are required to run a distributed routing protocol among themselves. The forwarding tables are sometimes are called routing tables.

3.2 Virtual circuit switching Before sending packets, a source needs to set-up a “connection” to the destination. States about the connection are stored in the switches on the path between the source and destination. An explicit connection set-up and an explicit connection tear-down are required. Unlike the circuit switching, packets from different connections are statistically multiplexed on a link. Need identifiers (Virtual Circuit Identifier, VCI) to distinguish them.

3.2 Virtual circuit switching An entry in the forwarding table usually consists of The VCI of an incoming packet, The interface from where the packet is received, The interface where the packet should be forwarded to, and The VCI for the outgoing packet. On a given link, packets belonging to a certain connection should be uniquely identified by a VCI.

3.2 Virtual circuit switching Switch 1 3 1 2 Switch 2 2 3 1 5 11 Host A 7 Switch 3 1 3 4 Host B 2

3.2.1 Routing problem in V. C. switching Two types of virtual circuits: Permanent Virtual Circuits (PVCs): A long-lived VC, usually pre-configured by administrators. Switched Virtual Circuits (SVCs): An on-demand VC, set-up by a source through signalling. Routing problem: How do the switches know how to forward a “connection set-up” message to the destination? The switches also need to run a distributed routing protocol for this purpose, e.g., P-NNI in ATM.

3.2.2 Properties of virtual circuit switching A source needs to wait for at least one RTT before sending packets. Same as the packet switching, virtual circuit switching also needs a routing protocol for the connection setup. Unlike the destination address for packet switching, the VCI has only local significance. The length of a VCI usually is a lot smaller than that of a complete destination address. A new connection needs to be set up in midst of link/node failures.

Source routing is a general approach to the problem of switching. Source routing can be implemented in switched LANs and IP networks. Source routing can be seen as one that is somewhere between packet switching and virtual circuit switching. Before sending packets, a source is required to find out the path to the destination. Unlike virtual circuit switching, the entire path information is put into each packet’s header.

3.3 Source routing

3.3 Source routing Header entering D C B A D C B A Ptr D C B A switch Header leaving A D C B D C B Ptr D C B A switch (a) (b) (c)

3.3.1 Properties of source routing A source needs to implement a mechanism to record the complete route to a destination. When finding a path to a destination, there is usually a cap on the maximum number of switches traversed. The source assumes that the path obtained before sending packets does not change during the packet delivery phase. When there are link/node failures, the packets may not be delivered to the destination.