Synchronization without contention

摘要: Conventional wisdom holds that contention due to busy-wait synchronization is a major obstacle scalability and acceptable performance in large shared-memory multiprocessors. We argue the contrary, present fast, simple algorithms for contention-free mutual exclusion, reader-writer control, barrier synchronization. These algorithms, based on widely available fetch-and-@ instructions, exploit local access shared memory avoid contention. compare our previous approaches both qualitative quantitative terms, presenting their Sequent Symmetry BBN Butterfly Our results highlight importance of memory, provide case against construction so-called "dance hall" machines, suggest special-purpose hardware support unlikely be cost effective machines with sequentially consistent memory.