Robust balancing in B-trees

Abstract

Robust balancing is a technique for maintaining generalized B-trees with a cumulative rebalancing cost that is asymptotically linear. It is especially significant in conjunction with fingers, which can make cumulative search cost linear. We define a new family of robust balancing algorithms which includes algorithms with sublinear rebalancing cost, extending previous work that has shown linear algorithms. An accounting technique is developed which gives strong bounds on rebalancing cost. We also consider robust balancing for B*-trees, a B-tree variant with improved space utilization.