2026/02/27 by Ilsun Chang · 1 voice
Computer Science · #Advanced Database Systems and Queries #Cloud Computing and Resource Management #Cost estimate #Executable #Execution time #Executor #Latency (audio) #Orchestration #Raw data #Robustness (evolution) #Software System Performance and Reliability #Unexpected events #cs.DB
paper · pdf · open access · doi:10.48550/arxiv.2603.02253
published in arXiv (Cornell University) (Cornell University)
openalex publication_date 2026/02/27 · arxiv published 2026/02/27 · arxiv updated 2026/02/27 · openalex created_date 2026/03/05 · openalex updated_date 2026/07/28
This paper analyzes execution instability in traditional cost-based database management systems (DBMS) and identifies a structural timing misalignment between optimization and execution stages that contributes to tail-latency amplification. Beyond estimation accuracy and raw execution throughput, we argue that decision timing and the availability of runtime signals materially affect robustness under uncertainty. In conventional DBMS architectures, the optimizer relies on historical statistics, the executor observes runtime data distributions and resource states, and accelerators impose up-front transfer costs and amortization constraints. This temporal asynchrony can lead to rigid early-bound decisions that fail under input-scale shifts or stale statistics. We propose a cross-layer decision timing orchestration framework that shifts final decision authority from the compile-time optimizer to the runtime executor via selective late binding of operator-level choices. A Unified Risk Signal (URS) integrates optimizer uncertainty, execution-time observations, and accelerator cost signals without collapsing them into a single static cost model. Experiments on a modified PostgreSQL prototype evaluate (i) input-scale shift, (ii) stale-statistics drift, and (iii) GPU offload break-even regimes using controlled microbenchmarks. The proposed orchestration improves execution stability, reducing P99 latency by up to 20x under severe estimation drift while maintaining comparable median latency.