摘要 |
PURPOSE: A method and an apparatus for maintaining a cache coherency of a multi-process system having a distributed shared memory is provided to enhance a performance of a system by referring multiple nodes at one time through a bus in a read and write mode in a cache coherency maintaining protocol method. CONSTITUTION: A directory(114) of each node is constructed in a pull map structure, and a cache state of a cache memory(112) of each node with respect to a data block is applied by 1-bit per node, and a cache state is indicated as one among a unshared, shared and dirty state. The directory indicates the entire node(n) and "u" represents a bit used. A source ID separates a start node for a certain transaction and is driven on a bus at the first period of the transaction. Only 1-bit is driven except for the concurrent transmission transaction. When the read operation is performed in a certain processor of the fist node(110), a node(initiator) which starts a read operation is the first node(110), it is checked that there is not a corresponding address in the memory(112) and request a read to the second node(120) having a corresponding address. The second node(120) which receives the read request checks the cache state of a corresponding address. If the third node(130) has a dirty block, a write back is requested with respect to a corresponding address. The information(source ID) of the first node(110) which starts the read operation is informed.
|