Tutorial10 Solution
Tutorial10 Solution
S. Wallentowitz
Why on-chip coherency is here to stay Milo M. K. Martin, Mark D. Hill, Daniel J. Sorin Communications of the ACM, July 2012
a)
A simple data bus (no pipelining) is used to connect the processor cores and the memory. What is the average bandwidth for each processor core assuming all cores generate consecutive memory accesses?
b)
The bus and memory allow for 4-beat bursts. How does the achievable bandwidth change?
c)
Develop a simple schematic sketch that shows the differences between a simple bus and a simple crossbar for the given scenario
d)
How does the bandwidth change for a crossbar? What is the limiting factor? In what different scenario can the performance be improved?
e)
How can a bi-directional ring improve the setup with respect to bandwidth and latency? Elaborate scenarios where a ring is advantageous and where it is disadvantageous
b)
Change the Channel-Dependency diagram of XY-routing so that it reflects west-first routing. Mark all forbidden turns and potentially add new routes.
c)
Check whether the depicted routing function is a valid turn model. If not, show the potential cycle.