I know we aren't talking about software methods for coherence, but am I correct in my intuition that the software methods would be slower than the hardware methods?
llcoolj
when looking into page-fault mechanism, i'm finding that page faults are raised by hardware? but is not a hardware-based solution?
ferozenaina
The page fault interrupt is raised by the (memory) hardware but the OS handles it. It occurs when the OS cannot find the (latest) data in virtual memory.
Also, in this case, we are talking about soft page faults, where the data is not up to date in the cache.
bpr
@Holladay, software-based methods could be faster than hardware-based approaches. First, the hardware can be simplified by not having to support coherence, saving latency and energy. Second, software can know in advance better than hardware where data will be needed.
I know we aren't talking about software methods for coherence, but am I correct in my intuition that the software methods would be slower than the hardware methods?
when looking into page-fault mechanism, i'm finding that page faults are raised by hardware? but is not a hardware-based solution?
The page fault interrupt is raised by the (memory) hardware but the OS handles it. It occurs when the OS cannot find the (latest) data in virtual memory.
Also, in this case, we are talking about soft page faults, where the data is not up to date in the cache.
@Holladay, software-based methods could be faster than hardware-based approaches. First, the hardware can be simplified by not having to support coherence, saving latency and energy. Second, software can know in advance better than hardware where data will be needed.
While hardware-based solutions are prevalent, a couple years ago, Why on-chip cache coherence is here to stay was again arguing this point.