Scholarlink

<p>Query optimization (QO) in distributed relational database management system (RDBMS) faces enduring challenges, such as high latency, computational overhead, and poor scalability exacerbated by cross-shard (CS) communication demands. Existing QO approaches often rely on on-chain mechanisms that require consensus among all relevant shards, limiting scalability and efficiency due to the overuse of cross-shard transactions (CSTs). This study introduces SGDSAC, a scalable self-governor learning framework based on distributional soft actor-critic (DSAC) that optimizes query execution plans (QEPs) in the blockchain-augmented distributed databases. Its core innovations include (1) a dynamic relational scoring mechanism for CST minimization and (2) a dual policy-value deep reinforcement learning (DRL) optimizer balancing QEP efficiency with exploratory plan generation. The framework incorporates an Algorand-inspired verifiable random function (VRF) and a decentralized service committee for distributed ledger management. Evaluations were examined on PostgreSQL using JOB, IMDB, and TPC-H benchmarks. JOB assesses performance under complex join operations; IMDB provides a real-world dataset to examine practical applicability, and TPC-H, with varying scale factors, evaluates scalability and workload handling. The experiments demonstrate SGDSAC’s superiority, achieving 25–50% lower latency, 12–60% cost reduction, 20–50% higher throughput, 4–8% higher speedup, and 8–37% Q-error improvements over state-of-the-art baselines.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

SGDSAC: a scalable self-governing DSAC-based learning framework for on-chain sharding and off-chain cloud databases

  • 期刊论文

  • 发表日期：

    2025年10月27日

查看全文
  • M. Khosravi,
  • S. H. Erfani,
  • M. Deypir

摘要

Query optimization (QO) in distributed relational database management system (RDBMS) faces enduring challenges, such as high latency, computational overhead, and poor scalability exacerbated by cross-shard (CS) communication demands. Existing QO approaches often rely on on-chain mechanisms that require consensus among all relevant shards, limiting scalability and efficiency due to the overuse of cross-shard transactions (CSTs). This study introduces SGDSAC, a scalable self-governor learning framework based on distributional soft actor-critic (DSAC) that optimizes query execution plans (QEPs) in the blockchain-augmented distributed databases. Its core innovations include (1) a dynamic relational scoring mechanism for CST minimization and (2) a dual policy-value deep reinforcement learning (DRL) optimizer balancing QEP efficiency with exploratory plan generation. The framework incorporates an Algorand-inspired verifiable random function (VRF) and a decentralized service committee for distributed ledger management. Evaluations were examined on PostgreSQL using JOB, IMDB, and TPC-H benchmarks. JOB assesses performance under complex join operations; IMDB provides a real-world dataset to examine practical applicability, and TPC-H, with varying scale factors, evaluates scalability and workload handling. The experiments demonstrate SGDSAC’s superiority, achieving 25–50% lower latency, 12–60% cost reduction, 20–50% higher throughput, 4–8% higher speedup, and 8–37% Q-error improvements over state-of-the-art baselines.