Pooled Memory Page Table for Faster Cluster Data Sharing

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing cluster systems require each node to repeatedly store page tables to access data in a memory node, limiting efficient data sharing among nodes.

Innovation Solution

Implement a pooled memory system with a shared pooled memory page table accessible by all nodes, allowing nodes to obtain physical addresses directly from this table without duplicating local page tables.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If each node stores its own page table to access memory node, then each node can independently access data, but memory bandwidth is wasted due to repeated page table storage

Engineering Contradiction:

Improveindependent data access capabilityVSAvoidmemory bandwidth utilization

Core Design Contradiction:

ReliabilityVSLoss of energy

Solution Approach 1:

The patent merges the page tables of multiple nodes into a single shared page table stored in the memory node. Instead of each node maintaining its own copy of the page table, all nodes reference the same page table located in the memory node's memory, eliminating redundant storage and optimizing memory bandwidth usage.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The shared page table in the memory node serves as a universal address translation resource for all compute nodes in the cluster. This single page table structure provides address translation services to multiple nodes simultaneously, making the memory node's memory serve multiple functions including storing both data and the shared page table.

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Speed

If each node stores page table locally, then access speed is fast, but system complexity increases due to duplicate page tables

Engineering Contradiction:

Improvedata access speedVSAvoidpage table management complexity

Core Design Contradiction:

SpeedVSDevice complexity

Solution Approach 1:

The patent combines multiple duplicate page tables into a single shared page table in the memory node, reducing system complexity. All nodes reference this unified page table structure, eliminating the need for each node to maintain separate page table copies and their associated management overhead.

Inventive Principle:

Principle #5Merging (Combining)

3Quantity of substance

If page table is shared in memory node, then storage efficiency improves, but access overhead increases

Engineering Contradiction:

Improvememory storage efficiencyVSAvoidpage table access overhead

Core Design Contradiction:

Quantity of substanceVSLoss of time

Solution Approach 1:

The patent extracts the page table from each individual node's memory and places it in the memory node's memory. This extraction eliminates redundant page table storage across nodes while the memory node's direct memory access capability minimizes the overhead of accessing the shared page table.

Inventive Principle:

Principle #2Taking out (Extraction)

Data Source

PatentUS20250390439A1Cluster system for sharing data using pooled memory and operating method thereof

Publication Date: 2025.12.25 SK HYNIX INC

AI summary

A cluster system may include a plurality of nodes each including a main memory, and a memory node including a pooled memory which is accessible by the plurality of nodes. The memory node may store, in the pooled memory, a pooled memory page table which indicates address translation information between a virtual address and a physical address of the pooled memory. The plurality of nodes may obtain a physical address of shared data stored in the pooled memory, from the pooled memory page table, and may access the shared data corresponding to the obtained physical address.