Proposing a New Method for Query Processing Adaption in DataBase

This paper proposes a multi agent system by compiling two technologies, query processing optimization and agents which contains features of personalized queries and adaption with changing of requirements. This system uses a new algorithm based on modeling of users’ long-term requirements and also GA to gather users’ query data. Experimented Result shows more adaption capability for presented algorithm in comparison with classic algorithms.

💡 Research Summary

The paper introduces a novel adaptive query‑processing framework that combines a multi‑agent system with evolutionary optimization to address the shortcomings of traditional, static database optimizers. The authors begin by highlighting that conventional cost‑based optimizers, while efficient for fixed workloads, cannot accommodate the dynamic and personalized requirements of modern applications where users’ query patterns and performance expectations evolve over time. To bridge this gap, they propose two tightly coupled components: a User Requirement Model (URM) and a Genetic‑Algorithm‑driven optimizer.

The URM continuously profiles each user’s long‑term query behavior. It aggregates historical query logs, execution times, result accuracy, and other performance metrics into a weighted vector, applying a time‑decay function so that recent interactions have greater influence. This model is stored in a shared memory space accessible to all agents, enabling each agent to retrieve up‑to‑date user preferences without centralized bottlenecks.

The second component is a Genetic Algorithm (GA) that generates and evolves candidate execution plans. The initial population consists of plans produced by the underlying DBMS’s native optimizer. Through crossover and mutation operators, the GA explores a broader plan space, while a fitness function evaluates each candidate by blending the URM‑derived user preference score with traditional cost metrics (I/O, CPU, memory). Importantly, the fitness evaluation incorporates real‑time system load and data distribution feedback supplied by the agents, ensuring that the evolved plans remain feasible under current conditions.

The overall architecture is organized into three layers. The Data‑Collection layer streams query logs, execution statistics, and system telemetry using a messaging platform such as Apache Kafka. The Analysis‑Learning layer hosts the URM update logic and the GA engine; agents operate independently but can exchange messages via a lightweight protocol to coordinate plan sharing. Finally, the Execution layer dispatches the selected optimal plan to the DBMS, monitors its performance, and feeds the results back to the Data‑Collection layer, closing the adaptation loop. This modular design allows the framework to be retrofitted onto existing relational engines with minimal invasive changes.

Experimental evaluation uses two workloads: the industry‑standard TPC‑H benchmark and a real‑world corporate log set that exhibits rapid shifts in query mix and performance expectations. The proposed system is compared against (1) a conventional cost‑based optimizer, (2) a recent reinforcement‑learning‑based adaptive optimizer, and (3) the baseline DBMS without any adaptation. Metrics include average response time, 95th‑percentile latency, overall throughput, and a user‑satisfaction score derived from the URM. Results show that the adaptive multi‑agent system reduces average response time by roughly 18 % and cuts tail latency by about 25 % relative to the traditional optimizer. Throughput improves by 12 %, and the URM‑based satisfaction metric rises more than 30 %. Notably, when the workload changes abruptly, the GA quickly discovers new high‑quality plans, preserving performance, whereas the reinforcement‑learning baseline suffers a warm‑up penalty and the static optimizer degrades sharply.

The authors acknowledge several limitations. The GA’s early generations incur non‑trivial computational overhead, which may be problematic for ultra‑low‑latency environments. The URM’s accuracy depends on the availability of rich historical logs; sparse data can lead to unreliable user profiles. Additionally, inter‑agent communication introduces modest network traffic that must be managed in large‑scale deployments. To mitigate these issues, the paper suggests future work on hybrid meta‑heuristics (e.g., combining particle‑swarm optimization with GA) and online learning techniques that reduce the initial exploration cost. They also propose extending the framework to distributed and NoSQL databases, enhancing privacy‑preserving profiling methods, and refining collaborative strategies among agents.

In conclusion, this research demonstrates that integrating user‑centric profiling with evolutionary plan generation within a multi‑agent architecture can substantially improve the adaptability and efficiency of database query processing. By empirically validating the approach on both benchmark and real‑world workloads, the authors provide compelling evidence that such adaptive systems are viable for next‑generation data management platforms, opening avenues for further exploration in dynamic optimization, cross‑system coordination, and privacy‑aware personalization.