I'm a little confused on why Processor 0 chooses to send a read request to the processor 1 directory, why not the processor 2 directory?
bojianh
I think the address points to somewhere in Processor 1's memory.
yangwu
The mapping can come from the address space directly if it is memory on different processor. but i am confused when it is cache.. since every processor can get that chuck of memory, how to determine which processor is the host?
acfeng
I think that the memory blocks corresponding to each processor have fixed addresses so one could locate the correct location.
bysreg
same question with anonymous, how does processor 0 know it needs to send the message to processor 1?
kayvonf
The mapping of address (in this case X) to a directory home node is fixed by the system. One simple solution is to design a simple hash function that sets this mapping. (hash(X) = home node id).
In this example, we asserted that node 1 is the home node for address X. Further discussion on this topic popped on up this slide.
I'm a little confused on why Processor 0 chooses to send a read request to the processor 1 directory, why not the processor 2 directory?
I think the address points to somewhere in Processor 1's memory.
The mapping can come from the address space directly if it is memory on different processor. but i am confused when it is cache.. since every processor can get that chuck of memory, how to determine which processor is the host?
I think that the memory blocks corresponding to each processor have fixed addresses so one could locate the correct location.
same question with anonymous, how does processor 0 know it needs to send the message to processor 1?
The mapping of address (in this case
X
) to a directory home node is fixed by the system. One simple solution is to design a simple hash function that sets this mapping. (hash(X) = home node id).In this example, we asserted that node 1 is the home node for address
X
. Further discussion on this topic popped on up this slide.