MUMBAI, India, June 30 -- Intellectual Property India has published a patent application (202621059144 A) filed by Kashika Wanchoo on May 09, 2026, for Distributed Database Memory Optimization System Using Predictive Allocation And Rdma Rebalancing.
Inventor includes Kashika Wanchoo.
The application for the patent was published on June 26, 2026, under issue no. 26/2026.
Abstract: The present invention discloses a distributed database query memory optimization system (100) integrating predictive memory allocation, physical memory reservation, real-time monitoring, and dynamic memory rebalancing. A Query Complexity Fingerprinting Engine (111) extracts a 150-dimensional feature vector from SQL queries and processes it using an XGBoost model comprising 500 decision trees to generate P50, P95, and P99 memory predictions in under 2 milliseconds. Predicted memory is reserved on Executor Nodes (120) using mlock()-based physical RAM locking to prevent memory exhaustion. A Memory Monitor (121) reads CPU hardware performance counters every 100 milliseconds to detect memory pressure. When utilization exceeds 90%, a Rebalancing Controller (113) initiates RDMA-based transfer of intermediate query data structures to donor nodes in 50 to 150 milliseconds, preserving in-progress execution. The invention reduces memory over-provisioning, mitigates query failures, and improves distributed query performance through a self-optimizing closed-loop architecture.
Disclaimer: Curated by HT Syndication.
