Can someone elaborate on the second point on message passing programs can execute on hardware that does not support system-wide loads and stores?
Aarow
I'm not 100% sure, but I believe that the second point is talking about the fact that each cpu needs a chunk of the memory to store buffer the message. The hardware shouldn't need to implement a system-wide load and store to execute message passing programs because messages can be local to a single cpu or between two cpus.
I don't think it's saying that the hardware can't support system-wide loads and stores, but it doesn't have to.
BigFish
MPI defines interfaces to pass messages between different processors or machines, which provides programmers the chance to write parallel code across a cluster. Unlike shared memory method, the cluster does not need to support system-wide load or store.
Can someone elaborate on the second point on message passing programs can execute on hardware that does not support system-wide loads and stores?
I'm not 100% sure, but I believe that the second point is talking about the fact that each cpu needs a chunk of the memory to store buffer the message. The hardware shouldn't need to implement a system-wide load and store to execute message passing programs because messages can be local to a single cpu or between two cpus.
I don't think it's saying that the hardware can't support system-wide loads and stores, but it doesn't have to.
MPI defines interfaces to pass messages between different processors or machines, which provides programmers the chance to write parallel code across a cluster. Unlike shared memory method, the cluster does not need to support system-wide load or store.