Scholarlink

This paper introduces a scalable HTAP architecture combining heterogeneous replication, dynamic partitioning, and batch processing to optimize OLTP throughput and OLAP query efficiency. By reducing cross-partition transactions and 2PC protocol overhead, it achieves 11x higher YCSB transactional throughput than existing systems while maintaining low latency. The design employs logical timestamp-based concurrency control and columnar sample datasets for efficient OLAP approximate queries, supported by epoch-based recovery for fault tolerance. Though batch processing introduces minor latency, the architecture balances scalability and consistency, with persistent logging ensuring resilience. Experiments validate significant throughput gains and analytical accuracy across workloads. Future work will enhance real-time data synchronization and extend support for machine learning/graph analytics. The solution addresses HTAP’s dual demands through integrated workload optimization, offering a robust foundation for data-intensive applications.

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

A Data Organization Architecture for Efficient Transaction Processing and Approximate AP Query in HTAP Systems

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Shengfei Shi,
  • Baiwei Zhao,
  • Xiaoxiao Xie,
  • Kaiqi Zhang,
  • Chao Yi

摘要

This paper introduces a scalable HTAP architecture combining heterogeneous replication, dynamic partitioning, and batch processing to optimize OLTP throughput and OLAP query efficiency. By reducing cross-partition transactions and 2PC protocol overhead, it achieves 11x higher YCSB transactional throughput than existing systems while maintaining low latency. The design employs logical timestamp-based concurrency control and columnar sample datasets for efficient OLAP approximate queries, supported by epoch-based recovery for fault tolerance. Though batch processing introduces minor latency, the architecture balances scalability and consistency, with persistent logging ensuring resilience. Experiments validate significant throughput gains and analytical accuracy across workloads. Future work will enhance real-time data synchronization and extend support for machine learning/graph analytics. The solution addresses HTAP’s dual demands through integrated workload optimization, offering a robust foundation for data-intensive applications.