Okay, that's a relief. I didn't fancy trying to explain all that in Spanish :-)
I'll use the example of modular switches but the principle is the same for fixed chassis switches as well.
In a modular switch, each module connects to the switch fabric of the chassis. The switch fabric is what is used to move packets to and from each module within the chassis.
Each module gets a certain amount of bandwidth to and from the chassis and this is usually fixed per slot ie. if there is unused bandwidth you cannot reuse it elsewhere
With Nexus switches you can increase the amount of bandwidth per slot by inserting more fabric modules but it is still fixed per slot ie. the amount of bandwidth may be more but each slot still gets a fixed amount.
So each module has a fixed amount of bandwidth to the chassis.
The other side of the equation are the end devices connected to the module ie. other switches./servers etc.
If you add up the total available bandwidth from all the ports and that is less than or equal to the modules connection to the switch fabric it is said to be non-blocking ie. all ports can run at wire speed.
If the amount of bandwidth from all ports is greater than the switch fabric connection of the module it is said to be oversubscribed.
So using a very basic example -
you have an 8 port 10Gbps module which equals a total of 80Gbps bandwidth from the ports.
If the module has an 80Gbps connection to the switch fabric then that is non-blocking.
if the module has a 40Gbps connection to the switch fabric then that is an oversubscribed architecture (2:1 oversubscription) because not all ports can run at 10Gbps concurrently without there being some blocking.
If you wanted to run in non-blocking mode you could only use 4 of the 8 ports.
So for example the 6500 switch is an oversubscribed architecture ie. the connections to the switch fabric do not match the potential traffic from the ports. How oversubscribed depends on a combination of the supervisor (which determines how much bandwidth is available per slot) and the module itself because different modules support different connections to the switch fabric.
In contrast, although I'm not familiar with all the Nexus switches, from what i have read most of them seem to be built around a non-blocking architecture which is why they are particularly suited to DC environments where the high throughput is needed.
One last point. When we talk about oversubscribed vs non-blocking we are talking about the potential bandwidth because you are not necessarily going to see all the ports on a module transmitting at peak rate at the same time.
But if you want to guarantee non-blocking under all circumstances then you would have to match the potential full bandwidth use from the ports to the switch fabric capacity.
Hope that's helped.
Jon
