How to Make a Multiprocessor Computer That Correctly Executes Multiprocess Programs

IEEE Transactions on Computers C-28 | , Vol 9: pp. 690-691

I forget what prompted me to be thinking about memory caching, but it occurred to me one day that multiprocessor synchronization algorithms assume that each processor accesses the same word in memory, but each processor actually accesses its own copy in its cache. It hardly required a triple-digit IQ to realize that this could cause problems. I suppose what made this paper worth reading was its simple, precise definition of sequential consistency as the required correctness condition. This was not the first paper about cache coherence. However, it is the early paper most often cited in the cache-coherence literature.