Adaptive BSTs for Single-Source and All-to-All Requests: Algorithms and Lower Bounds

TL;DR

This paper introduces a unified framework for adaptive binary search trees tailored for single-source and all-to-all request models, providing algorithms, bounds, and a theoretical analysis of their performance in both offline and online settings.

Contribution

It presents new offline algorithms with proven upper bounds and extends them to online strategies, along with lower bounds on their competitive ratios, for adaptive BSTs in networked systems.

Findings

Proposed offline algorithms with cost bounds for both models

Extended algorithms to online adaptive BST strategies

Established lower bounds on online algorithm competitiveness

Abstract

Adaptive binary search trees are a fundamental data structure for organizing hierarchical information. Their ability to dynamically adjust to access patterns makes them particularly valuable for building responsive and efficient networked and distributed systems. We present a unified framework for adaptive binary search trees with fixed restructuring cost, analyzed under two models: the single-source model, where the cost of querying a node is proportional to its distance from a fixed source, and the all-to-all model, where the cost of serving a request depends on the distance between the source and destination nodes. We propose an offline algorithm for the single-source model and extend it to the all-to-all model. For both models, we prove upper bounds on the cost incurred by our algorithms. Furthermore, we show the existence of input sequences for which any offline algorithm must incur a cost comparable to ours. In the online setting, we develop a general mathematical framework for deterministic online adaptive binary search trees and propose a deterministic online strategy for the single-source case, which naturally extends to the all-to-all model. We also establish lower bounds on the competitive ratio of any deterministic online algorithm, highlighting fundamental limitations of online adaptivity.