A secondary cache control system for a computer system is disclosed. The system is utilized advantageously to reduce the cost of the SRAM while not degrading the overall performance of the CPU associated with the computer. The system latches the data from the CPU until the CPU hits a "dead time". When this dead time occurs, the data is written into the SRAM. By writing to the SRAM at this time the performance of the computer system is not degraded and the cost of the SRAM is significantly reduced.

A non-shared system with respect to an outside and an inside cache in a multi-processor system has multi-layer hierarchical cache. An invalidation address on a main memory address bus 31 in company with the rewriting of a main memory 30 is transmitted via a first and a second path 35, 36 to inside caches 11,21 so as to invalidate these inside caches 11, 21. The invalidation address is transmitted to the outside caches 12, 22 via the bidirectional connections between the main memory address bus 31 and outside caches 12, 22 so as to invalidate these outside caches 12, 22. It is extremely improbable for the address of access for writing to the transmitted to the main memory address bus 31 because the outside caches 12, 22 are operated in either copy back or write once system. As a result, even though the invalidation address is transmitted to the outside caches 12, 22 via the bidirectional connections between the main memory address bus 31 and the outside caches 12, 22, it is extremely improbably for the address of access for writing to conflict with a signal on the bidirectional connection.

A two domain digital network with each domain having its own system bus and its own bus exchange module permits Write operation addresses to be passed between domains. Each bus exchange module provides a match filter which prevents the passage from one bus to the other bus of a duplicate Write operation (OP) address which has already been transferred, thus relieving the busses of excess traffic when a duplicate Write OP address is being sent to a cache memory for an invalidation operation. A Read operation will nullify the match filter to then allow passage of each incoming Write OP invalidation address to the snoop invalidation queue, but prevent the passage of a subsequent duplicate Write OP address, so long as the read OP is ongoing.

A state machine system is used to control a cache controller in a network involving the operations of a processor having a store-through cache and operations involving an invalidation queue which is filled by a spy module which monitors dual system busses to select addresses of words which appear for write operations.

A FIFO invalidation queue for address words, from a spy module, are held for subsequent invalidation operations to a cache memory. The FIFO queue is programmably organized to indicate when it is almost full in which case it will switch to a priority operation which will give priority to invalidation cycles in the cache over the priority of the processor's cache access. When the FIFO queue indicates that it is almost empty, then the priority of the cache access by the processor is re-established as it was in normal conditions. The system operates concurrently in a self-regulating manner to load and unload addresses into the FIFO queue while also giving priority to flushing out the queue with invalidation cycles when preset upper limits are reached.