The reason we cut the network in half for bisection bandwidth is to get the worst case bisection which results in maximum bandwidth across the severed link as there is half the nodes on each side.
yeezus98
Is having half the nodes as severed links necessarily the worst case bisection?
MangoSister
When talking about bisection bandwidth, are we assuming the cut is the best possible cut to get largest bisection bandwidth (since there are many ways to cut)?
ferozenaina
For bisection bandwidth, you make the cut so that there are equal number of nodes on each side which have to communicate.
A property of a desirable interconnect is increasing bisection bandwidth with increasing number of nodes. This means the network scales well.
The reason we cut the network in half for bisection bandwidth is to get the worst case bisection which results in maximum bandwidth across the severed link as there is half the nodes on each side.
Is having half the nodes as severed links necessarily the worst case bisection?
When talking about bisection bandwidth, are we assuming the cut is the best possible cut to get largest bisection bandwidth (since there are many ways to cut)?
For bisection bandwidth, you make the cut so that there are equal number of nodes on each side which have to communicate.
A property of a desirable interconnect is increasing bisection bandwidth with increasing number of nodes. This means the network scales well.