Centralized MAC and SwitchingENEE 426 | Communication Networks | Spring 2008 Lecture 5Multiple Access• Infrastructure: – Control access to uplink• Distributed: – Control access to shared medium– Distributed networks are half-duplex– No central controller -> users = networkENEE 426 | Communication Networks | Spring 2008 Lecture 5Ring Topologies• Early centralized approach called token ring• Devices could only transmit if they had the “token”• Passed the roll of transmitter around the networkENEE 426 | Communication Networks | Spring 2008 Lecture 5Token Ring• Token frame (no payload) circulates network until a device needs to transmit• Transmitting device receives token, attaches payload, transmits• Packet circulates around the network• When it returns to the originating device, the payload is removed and the token is forwarded onENEE 426 | Communication Networks | Spring 2008 Lecture 5Token Ring Performance• Deterministic– Same amount of time to transmit every packet– Finite maximum amount of time between transmissions• Compared to CSMA/CD– Better overall performance– No chance of collisions– But more overhead to maintain tokens• Switched Ethernet put Token Ring out of businessENEE 426 | Communication Networks | Spring 2008 Lecture 5Quality of Service• Each node has a priority• To transmit– If received token/data frame has higher priority already, wait for next token– Otherwise change priority to match yours– If empty token arrives with your priority, transmit– When token returns, reset priority to original valueENEE 426 | Communication Networks | Spring 2008 Lecture 5FDDI and RPR• FDDI was fiber version of token ring• Initial interest, but died out• RPR is ring-like topology for SONET/SDH• Used in some MANsENEE 426 | Communication Networks | Spring 2008 Lecture 5Centralized MAC• Centralized star topologies• All devices communicate with a single relay node rather than with each other– Wireless: base station– Ethernet: switch• Wired networks: each user has own channel• Wireless: shared channelsENEE 426 | Communication Networks | Spring 2008 Lecture 5Channel Assignments• Multiple Approaches:– Frequency Division Multiple Access• Different users use different frequencies– Time Division Multiple Access• Different users transmit at different times– Hybrid (OFDMA); Other (CDMA)ENEE 426 | Communication Networks | Spring 2008 Lecture 5Comparison• FDMA – Typically requires fixed-width channels– Good for circuit switched networks– Too few users, wasted capacity– Too many users, some users receive no service• TDMA– Size of time blocks typically fixed– Vary who gets assigned to which blocks when– Users get 1/n of the available bandwidth– If underutilized, others can increase speed• CDMA– Power levels fixed– More users means more interference, so more coding necessary, reducing overall rateENEE 426 | Communication Networks | Spring 2008 Lecture 5Switched Networks• Virtual Circuits– Connection-oriented communications on packet-switched network– Network preconfigured with identifiers for source-destination pairs– Virtual Circuit IDs (VCIs)– Packets transmitted with virtual circuit ID in the header– Devices map this VCI to physical interfaces as the packet transits to properly route it• Good for QoS– Can guarantee enough capacity to support all the VCs at particular rate• Bad for dynamic networks– Overhead associated with setup/teardown of VCsENEE 426 | Communication Networks | Spring 2008 Lecture 5Switched Networks• Source Routing– Each packet includes a list of each intermediate switch, so switch knows how to transmit packet through the network– Avoids setup overhead, but no QoS guarantees– Dumb switches, but smart hostsENEE 426 | Communication Networks | Spring 2008 Lecture 5Switched Networks• Packet Switching– Dumb hosts, but smart network– Hosts address packets, network figures out how to deliver them• Bridges/Switches– Learn location of MAC addresses through bridging tables• Associate MAC addresses with certain ports• Bridging loops?ENEE 426 | Communication Networks | Spring 2008 Lecture 5Bridging/Switching LoopsENEE 426 | Communication Networks | Spring 2008 Lecture 5Spanning Tree Protocol• Used to generate loop-free switching topologies– Prevent broadcast storms– Prevent bridging table failure• Basic goal– Compute a TREE that SPANS the entire network without introducing any loops– Not all trees are optimal• May take advantage of low-rate links• Introduce bottlenecks– Protocols try to compute minimum weight spanning treeENEE 426 | Communication Networks | Spring 2008 Lecture 5Spanning Tree Protocol• Each bridge has unique ID and priority number• Basic Idea– Elect a root bridge (lowest priority)– Each bridge computes least-cost path from itself to the root bridge (provisions for breaking ties)– Cost is function of link speeds• Protocol– STP has unique MAC address (broadcast to other switches)– Every two seconds STP messages are exchanged• Weight to believed root bridge• Notification of topology changeENEE 426 | Communication Networks | Spring 2008 Lecture 5STP