I found a more straight-forward and intuitive definition of blocking network: In telecommunications, a network is blocking if it has fewer transmission paths than would be required if all users were to communicate simultaneously. (Note: Blocking networks are used because not all users require service simultaneously. Certain statistical distributions apply to the patterns of user demand.)
jaguar
In this case the blocking occurs because for certain combinations of messages, there do not exist any non-conflicting paths.
acfeng
If we wanted to avoid blocking when sending multiple requests, is there any way to figure out beforehand to understand whether or not these requests would be blocking?
bysreg
Just in the previous slide, non-blocking network is defined as a network which any pairing of nodes is possible. I think this network fits that definition yet in this slide it is called as blocking because there is conflict with some simultaneous messages. Which one is correct?
haibinl
@bysreg Yes we should consider whether there's conflict in sending messages to any pairs.
RX
How to theoretically veriy that if a network typology is blocking or not?
If a nonblocking network is a network that allows all to all communications happened at the same time, isn't the network must be fully connected?
traveling_saleswoman
This is a question that also came up in the review session yesterday: is there an example of a non-blocking network that is not fully connected? I don't remember the TAs giving a conclusive answer on this.
a
@traveling_saleswoman Not sure if this fully answers the question, but it is possible to make a non-blocking network with fewer than $n^2$ wires that a crossbar would use. See nonblocking minimal spanning switches. Basically, you can use several lower-degree crossbars in three layers to simulate a single higher-degree crossbar, but you have to add logic to reroute a connection in real time if a conflict comes up.
I found a more straight-forward and intuitive definition of blocking network: In telecommunications, a network is blocking if it has fewer transmission paths than would be required if all users were to communicate simultaneously. (Note: Blocking networks are used because not all users require service simultaneously. Certain statistical distributions apply to the patterns of user demand.)
In this case the blocking occurs because for certain combinations of messages, there do not exist any non-conflicting paths.
If we wanted to avoid blocking when sending multiple requests, is there any way to figure out beforehand to understand whether or not these requests would be blocking?
Just in the previous slide, non-blocking network is defined as a network which any pairing of nodes is possible. I think this network fits that definition yet in this slide it is called as blocking because there is conflict with some simultaneous messages. Which one is correct?
@bysreg Yes we should consider whether there's conflict in sending messages to any pairs.
How to theoretically veriy that if a network typology is blocking or not?
If a nonblocking network is a network that allows all to all communications happened at the same time, isn't the network must be fully connected?
This is a question that also came up in the review session yesterday: is there an example of a non-blocking network that is not fully connected? I don't remember the TAs giving a conclusive answer on this.
@traveling_saleswoman Not sure if this fully answers the question, but it is possible to make a non-blocking network with fewer than $n^2$ wires that a crossbar would use. See nonblocking minimal spanning switches. Basically, you can use several lower-degree crossbars in three layers to simulate a single higher-degree crossbar, but you have to add logic to reroute a connection in real time if a conflict comes up.