The MSI diagram is symmetrical (each "to" edge has a corresponding "from" edge), but the MESI diagram isn't. In particular, there is no direct M->E or S->E transition.
The M->E transition wouldn't make sense to have because the only two reasons to leave the M state are when someone else plans to read or write. In either case, the first owner wouldn't have exclusive access to the data.
The S->E transition also wouldn't make sense. At the moment you enter the S state, you're guaranteed that someone else is also in the S state (otherwise you would go to the E state). But you don't know if/when the shared owners drop to the I state after an eviction. So to move to the E state, you would have to ask if anyone is still sharing that data. This defeats the whole purpose of the E state though, which is to reduce communication.
This comment was marked helpful 0 times.
vrkrishn
In addition, there the advantage of the MESI protocol lies in the common operation of reading some data from memory and then repeatedly reading that data and writing to that data. If our processor is at the exclusive state, no signaling has to be done to move the data into the modify state or read data in the exclusive state. However, in the MSI protocol, reading data and then moving to the modify state will require the processor to send a BusRdx to every other processor even though no other processor will respond.
However, an offset of this is that data loads must be done by sending a signal to the other processors and waiting for a result from those processors. This is in contrast to the MSI protocol where no processor needs to hear responses from other processors after sending a signal.
The MSI diagram is symmetrical (each "to" edge has a corresponding "from" edge), but the MESI diagram isn't. In particular, there is no direct M->E or S->E transition.
The M->E transition wouldn't make sense to have because the only two reasons to leave the M state are when someone else plans to read or write. In either case, the first owner wouldn't have exclusive access to the data.
The S->E transition also wouldn't make sense. At the moment you enter the S state, you're guaranteed that someone else is also in the S state (otherwise you would go to the E state). But you don't know if/when the shared owners drop to the I state after an eviction. So to move to the E state, you would have to ask if anyone is still sharing that data. This defeats the whole purpose of the E state though, which is to reduce communication.
This comment was marked helpful 0 times.
In addition, there the advantage of the MESI protocol lies in the common operation of reading some data from memory and then repeatedly reading that data and writing to that data. If our processor is at the exclusive state, no signaling has to be done to move the data into the modify state or read data in the exclusive state. However, in the MSI protocol, reading data and then moving to the modify state will require the processor to send a BusRdx to every other processor even though no other processor will respond.
However, an offset of this is that data loads must be done by sending a signal to the other processors and waiting for a result from those processors. This is in contrast to the MSI protocol where no processor needs to hear responses from other processors after sending a signal.
This comment was marked helpful 1 times.