Question: How does MESIF work? Specifically, how is the cache that supplies data in the event of a cache miss determined in MESIF? (BTW: Wikipedia has a nice description)
Question: How does MOESI work? (again, see Wikipedia)
This comment was marked helpful 0 times.
tliao
As mentioned above, only one cache can be in the F state for a given cache line. The way this works is that a LIFO (last-in, first-out) policy is used to determine which cache gets to be the "data provider" for the line; the most recent cache to request the line gets maintains it in the F state, while the previous holder transitions into the S state.
This comment was marked helpful 0 times.
GG
In MOESI protocol, transition from M to S requires no flush. Instead, it defers the flush and transits into the Owned(O) state, and supply the modified data directly to the other processor. Other processors reading this line should be in S state.
If another processor tries to write to the line, it may tell the processors in O and S states to update their copies with the new contents and transit to S state, marking its own copy in the O state. This involves a transition of the Ownership of this line.
This comment was marked helpful 0 times.
kayvonf
The thinking in MESIF is that the last cache to request a line is the most likely cache to have the line well into the future. (The least likely cache to evict the line soon.) For simplicity, if a line in the F state is evicted, the next processor trying to load the line will retrieve the data from memory (even though there might still be other caches with the line in the S state).
To avoid this memory transfer, a more complex coherence protocol might instead try to promote a cache with a line in the S state to the F state in the event the F-state line is evicted. This would require the caches to coordinate to choose the new owner. It sounds complicated. As you are learning in this class, simplicity typically wins.
Question: How does MESIF work? Specifically, how is the cache that supplies data in the event of a cache miss determined in MESIF? (BTW: Wikipedia has a nice description)
Question: How does MOESI work? (again, see Wikipedia)
This comment was marked helpful 0 times.
As mentioned above, only one cache can be in the F state for a given cache line. The way this works is that a LIFO (last-in, first-out) policy is used to determine which cache gets to be the "data provider" for the line; the most recent cache to request the line gets maintains it in the F state, while the previous holder transitions into the S state.
This comment was marked helpful 0 times.
In MOESI protocol, transition from M to S requires no flush. Instead, it defers the flush and transits into the Owned(O) state, and supply the modified data directly to the other processor. Other processors reading this line should be in S state.
If another processor tries to write to the line, it may tell the processors in O and S states to update their copies with the new contents and transit to S state, marking its own copy in the O state. This involves a transition of the Ownership of this line.
This comment was marked helpful 0 times.
The thinking in MESIF is that the last cache to request a line is the most likely cache to have the line well into the future. (The least likely cache to evict the line soon.) For simplicity, if a line in the F state is evicted, the next processor trying to load the line will retrieve the data from memory (even though there might still be other caches with the line in the S state).
To avoid this memory transfer, a more complex coherence protocol might instead try to promote a cache with a line in the S state to the F state in the event the F-state line is evicted. This would require the caches to coordinate to choose the new owner. It sounds complicated. As you are learning in this class, simplicity typically wins.
This comment was marked helpful 0 times.