摘要 |
<p>A method and apparatus for minimizing memory read latency in a digital computer system which includes a system control unit (SCU) with an associated main memory, and a plurality of independent central processing units (CPU's), each CPU having an associated writeback cache. Data requests from a requesting CPU start a normal read cycle in the SCU main memory, and the data read requests concurrently begin a search for the location of any modified version of the requested data in the SCU data storage location directory, which stores the location of modified data located in both the SCU main memory and the caches of all the CPU's. The SCU main memory read cycle is aborted if a modified version of the requested data is found in any of the CPU caches, and the modified version of the requested data is then passed from the located CPU cache and through the SCU to the requesting CPU as read data.</p> |