Cause of the practical decline of the horizontal growth scheme

Ascertain whether the observed discrepancy between the theoretical optimality of the horizontal growth scheme for LSM-trees and its practical decline in adoption is caused by two specific limitations: (1) the horizontal scheme’s limited capability to handle update-heavy workloads due to its design only for the leveling merge policy and lack of effective tiering support, and (2) the gap between theory and practice arising from the horizontal scheme’s reliance on full compaction versus the file-level partial compaction commonly used in industrial systems.

Background

The paper analyzes two major LSM-tree growth schemes—vertical and horizontal. Although prior theory (e.g., Bentley and Saxe) suggests that horizontal growth can yield optimal read-write trade-offs under leveling, most modern systems favor vertical growth. The authors highlight this tension and explore whether practical limitations explain the divergence.

They identify two potential factors that might undermine horizontal growth in practice: difficulties with update-heavy workloads due to lack of tiering integration, and a mismatch with industrial practice where partial compaction is preferred over full compaction. The paper proposes horizontal-tiering and a hybrid scheme (Vertiorizon) to address these limitations, but the causal explanation for the adoption discrepancy is articulated as a conjecture rather than a settled result.