[NTLUG:Discuss] Ethernet Switches

[David Neeley]

At 11:55 AM 10/8/2001 -0500, you wrote:
>Eric Schnoebelen wrote:
>
> > Dennis Myhand writes:
> > - No.  It is a switch if it uses memory to store the packets before
> > - forwarding them and transistors (or some other suitable electronic
> > - devices) to make the connections between ports.  It is an hub if
> > - the connection between ports is always there and the packet from
> > - one machine on the hub goes to ALL machines on the hub.
> >
> >         Very close.  The packet doesn't have to be stored before
> > forwarding (and the mid-range and higher products don't).
> > Basically, in a switch, a port never sees a packet not addressed
> > for that port (via the ethernet MAC address.)
>
>Sort of.  Cut-through mode on a Cisco switch reads the first 64 bytes of a
>packet before moving it on its way.  Store-and-forward reads the packet into
>the memory of the switch before sending it on its way.

Classic store-and-forward in Ethernet networks was most often the case in a 
bridge. As I understand it, a switch can replace both hub and bridge in the 
architecture. It should be noted that a switch or an enterprise hub can 
both combine various types of transports (such as 10B2 and 10BT, as can a 
bridge. However, a bridge and a switch can also translate between various 
protocols, perhaps Ethernet and X25, for example...or so I *think* I 
understand! (Feel free to correct me if I'm wrong on any of this!)

Unless I am misreading, apparently the status of having the packet in the 
switch memory momentarily is not considered a "store-and-forward" event. 
Consider the definition in Newton's Telecom Dictionary: ""To qualify as an 
Ethernet Switch, the device must: Be capable of switching packets from one 
Ethernet segment to another "on-the-fly;" Avoid using slower 
store-and-forward technologies to route packets from one segment to 
another; Exhibit very low port-to-port latency (the elapsed time between 
receiving and transmitting a LAN packet is measured in 10s of microseconds, 
not 100s); Offer a busless, scaleable architecture that increases network 
carrying capacity as switched connections are added..."

By contrast, a hub contains a backplane with multiple busses, designed to 
carry data between any two ports without impeding dataflow between others.

In looking at these things, I also looked at some related definitions that 
have always puzzled me but which I have never actually studied to compare them:

Managed hub...Allows remote administration, troubleshooting and monitoring. 
Management tools typically include five categories: accounting, 
configuration, performance, fault, and security management. Most support 
the Simple Network Management Protocol.

Passive hub     Only does the connection.

Active hub      Amplifies the signal to allow longer cable runs, successful 
uplinks to join hubs together, etc.

>Because full can transmit and receive simultaneously, it is twice as fast.

It is "twice as fast" only if the information is relatively equal in moving 
each way. In a typical client/server architecture, that may be assuming a 
great deal more than is actually the case. However, there is no doubt the 
potential is indeed double or more. In fact, it may be that it's more than 
double in real world terms because of the time outs as the parties detect a 
collision. With a full duplex connection, there don't need to be these 
time-outs.

David