At 09:20 AM 3/14/02, John Green wrote: >a node connects to a switch and switch in turn >connects to all other nodes. hence in effect when a >node transmits it is the only one transmitting on that >wire and hence gets the full bandwidth in its transmit >wire (eg in 10BaseT). (csma/cd not applicable >here,....right ?
Actually, CSMA/CD still comes into play if the port is set to half duplex. The node gets full access to its transmit pair, but unless you are using full duplex, the node may experience what it thinks are collisions. If the node receives on its receive pair while sending on its transmit pair, it considers this a collision event. This is sort of an artifact of how CSMA/CD works and can be avoided by enabling full duplex. Full duplex is appropriate in almost all cases when only one device connects to a switch port. (I say in almost all cases, because sometimes you may not want to use it. For example, if a server is heavily loaded, but medium contention has limited the rate at which transactions are hitting it, upgrading to a full-duplex switching port could sufficiently overload the server to crash it.) > because it is the only node >transmitting on it transmit wire connecting to the >switch) > >But what if two or more nodes are trying to send >packets (rather frames) to one particular node. >say >two frames from two different nodes, destined for node >A arrive in the switch and now how does the switch >send the frame (frames), or which frame would it send >to node A ? and what happens to the other frame ? is >it discarded by the switch or is it quequed in the >memory and is sent next. The second frame is queued and not dropped, assuming enough memory is available. >how does it work ? > >csma/cd would apply here and bandwidth would have to >be shared in such a case ??? This isn't CSMA/CD, but bandwidth and other resources in the switch are definitely shared, so your thinking is right on. A switch is like a big Japanese Pachenko machine with frames going in all directions at once and getting held up in some cases and popping out later. Switch vendors distinguish themselves by designing ingenious switching fabrics to handle the issue that you mentioned and similar issues. The queuing has to be handled carefully to avoid "head of the link blocking." Consider an architecture where a frame is buffered on input because the output port is busy. Then another frame comes in. Regardless of where it's destined, it's going to get held up because of the frame already in the input queue! That's called head of the line blocking. It can be avoided by shipping off the input packet to some shared buffer or by using virtual output queuing. Telecom companies have worked on switching architectures for many years. A crossbar switch was invented by Reynolds for Western Electric in 1913. Many modern switches still use a crossbar architecture. If you search in Google on "crossbar switching," you'll find some interesting but esoteric stuff on this topic. A good book that covers the topic of switching architectures is: "Routing and Switching: Time of Convergence?" by Rita Puzmanova. Priscilla >__________________________________________________ >Do You Yahoo!? >Yahoo! Sports - live college hoops coverage >http://sports.yahoo.com/ ________________________ Priscilla Oppenheimer http://www.priscilla.com Message Posted at: http://www.groupstudy.com/form/read.php?f=7&i=38320&t=38227 -------------------------------------------------- FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]