Optimizing transaction throughput in databases via an intelligent scheduler
Samuel Kaspi · 2002
The author investigates the effect that a discriminating scheduler, which has a priori knowledge of the probable effect that each transaction would have on data contention, could have on throughput. The author finds that in those systems where both low contention and high contention transactions co-exist in significant proportions, a discriminating scheduler can increase throughput to near those levels which could be attained if no contention existed at all. The author investigates such a scheduler under a mix of transaction types and finds that in most cases, its effect on throughput is very significant.