W ork L oad W eaver : Realistic Synthetic Query Workloads for Adaptive Database Systems
Year: 2026
Type: conference-paper
Abstract: Self-driving data systems require realistic query workloads to train and evaluate adaptive components, but production traces are rarely shareable, and template-based benchmarks miss key dynamics such as temporal variation, session-driven dependencies, refinements, and workload shift. We present WorkLoadWeaver, a benchmark-grounded pipeline that constructs a workload specification from a set of query templates and optional domain assumptions. The specification models personas and temporal profiles, persona-conditioned query transitions with within-template refinements, and event-driven perturbations. WorkLoadWeaver additionally supports varying user participation, ranging from fully automated spec construction to human checkpoints that provide structured summaries and allow edits and constraints.
Institution: University of California, Irvine
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Funder: National Science Foundation
Awards: SCC-CIVIC-FA Track B: CaReDeX: Enabling Disaster Resilience in Aging Communities via a Secure Data Exchange, EAGER: LLM based Enrichment for Data Analytics, III: Small: EnrichDB - Supporting Enrichment in Database Systems, G56477762, IUCRC: UCI: Planning Grant: Proposal Planning Grant: Center for Smart Space Research (CSSR), Collaborative Research: CPS: NSF-JST: Enabling Human-Centered Digital Twins for Community Resilience, Collaborative Research: SaTC: CORE: Medium: Compliance as a Service (CaSe): A Reflective Approach to Enforcing User Privacy Regulations, SCC-IRG-Track2: Creating an Extensible Data Exchange and Analytics Sandbox for Smart Water infrastructures
OpenAlex ID: W7168564345