Peter Nachtwey said:
I know that there are often 5 of our products per machine center and each of them is producing status data at a very high rate. A router must isolate these packets to the local machine centers otherwise the whole network will be overloaded and there will be delays due to collisions.
This is only true if you use lots of hubs or archaic ethernet technologies. A modern switched ethernet network can have all the local 'machine centers' on the same IP subnet without a problem.
I think you are misunderstanding collisions. At the lowest level of the TCP/IP protocol is the hardware level. This is where all the nitty gritty details such as voltage level and signal encoding happen. Ethernet exists partially at this level.
In it's original form, ethernet used common media for all nodes on the network. A collision would happen when two nodes on the network tried to simeotaneously submit messages across the shared medium rendering both messages useless. The collision would be detected and both nodes would use an algorithm that determined a random wait period before trying to re-transmit. This worked fine as long as netowrk traffic wasn't heavy or there weren't too many nodes on the network. During heavy usage periods, ethernet was prone to collision storms where lots of nodes were tyring to transmit at the same time, but very little data was being transferred. All the bandwidth was wasted on collisions.
Devices known as repeaters were used to regenerate a signal on an ethernet network when the shared medium became too long to successfuly transmit the message. Later hubs, also known as multi-port repeaters were used to plug a bunch of nodes into the same shared medium. Hubs and repeaters are layer 1 devices that simply repeat the signal. So every device on a hub recieves every message sent out by any node. The node must decide if the message was intended for it or not.
After ethernet was in use for a while, large shared medium networks were shown to be painfully slow because 'collision storms' during high network activity could bring an ethernet network to it's knees.
This introduced the concept of a bridge. A bridge was a two port device that was slightly more intelligent than a hub. A bridge could be inserted in between two hubs. It would analyze messages sent out from each hub (which could be connected to several nodes) and decide if the message was inteded for a node on the same port, or a node on the other port. It would only regenerate the message if the message was intended for a node on the other 'side of the bridge.' This created seperate collision domains and alllowed a way to be segment large networks that were succeptible to collison storms. Bridges are considered layer 2 devices that use the MAC or hardware encoded address of a node to determine which side of the bridge the message was intended for.
Later, the concept of a multi-port bridge, commonly known as a switch, was introduced. This greatly increased network performance. Each node attached to a switch was it's own collision domain, so it eliminated the possibility of collisions as long as only one node was attached to one port on the switch. This paved the way for full duplex 100 mbps ethernet.
This all worked fine for local networks, but what happens when one local network needs to be connected to another netowrk? This is where routers come in to play. Routers are devices that are intended to pass messages between different networks. They use the network address (IP address in TCP/IP) to determine if the message should be sent to another network or ignored and left to the local network to handle.
Modern managed high end switches can act as 'routers' on local networks by segmenting networks into Virtual LANs or VLANs. This way logical grouping of nodes can be accomplished as opposed to physical grouping of nodes.
and anyway, it's obvious that i've studied this **** too much and i'm going to shut up about it now.