The last 3 bytes can be a destination that the vendor can assign per device
From Hubs to Switches/Bridges
Hubs are physical layer repeaters that broadcast the input to all endpoints on the signal -> can lead to collisions
high bandwidth cables resulted in CD support for small distances only bc of pipe size limit of 64 bytes
so, change to switches: point-to-point, multi-input, multi-output endpoint, but only 1-to-1 signals, without collisions -> separate bridge for each connection
switches store and buffer frames and relay entire frames, not bytes
Ethernet hub limitation on bandwidth and distance
suppose multiple stations and 2 separate LANs/ethernets, how to interconnect ethernets and double bandwidth, distance, and stations
cant use routers because routers at the time (1980s) were dependent on protocol: Xerox or Dack?
using a brige/switch that buffers data, if the hub doesn’t know the dest mac, just flood it and send to all stations on eth2 -> doubles distance and stations but not bandwith
instead, have a DB for mac addrs and switch port ids (eth1 or eth2 line in/out) -> build the DB on the switch using frames src and dest macs and eth line id then you can know where the data is going and whether its going up or down (like defaultdict)
keys -> eth line/id, values are mac addrs
flood if addr not in db and update db
remove least recently used (or some other metric) -> rarely required as db is large enough
only condition is non-cyclic topology -> any tree topology is ok
concurrency is not an issue because frames are buffered, so switches wait if there is data on the line with the same mac dest as the buffered data (e.g., multiple switches on a shared ether)
timer for buffering, if timer expires, flood -> something on the order of 10 mins
code
Terms
transparency - switches must be transparent to stations, stations should not care whether the switch exists or have knowledge abt it
promiscuous receive - switches buffer from all stations regardless of src
flooding - packets forwarded to all stations on the line and picked up by correct mac
filtering - switches can filter packets based on headers to filter forwarding or buffering
Design
Routing - Spanning Tree
every bridge enables parent port and
Autoconfiguration (Multicast)
based on multi-cast addressing
if MSB of MAC addr is 1 -> multi-cast addr (if multiple 1s in MSBs -> broadcast addr)
multi-cast - sender broadcasts signal to a subset of stations (instead of single)
multiple like-kind servers (e.g., file servers) and 1 responds (note requires src mac addr in eth header)
vendors get unicast and multicast addresses (they get assigned 2, 3 byte venddor identifiers, one with leading 0 bit for unicast and one with leading 1 for multicast addrs)
