Lock-granularity analysis tools in SVR4/MP

The design of the SVR4/MP multiprocessing system with fine-grained locking is discussed. SVR4/MP scales efficiently to eight processors under a database transaction processing workload and typical multiuser benchmarks. In developing SVR4/MP, initial lock placement was deemphasized and rapid, iterative tuning of lock placement was the focus. Sophisticated tools were used to perform iterative overall performance and scalability tuning of the operating system. The tools developed to determine lock granularity quantitatively are described and an example of their use to optimize overall system efficiency is presented.<
>