Rank Aggregation for Memory Expansion

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing memory access systems face challenges in expanding memory capacity without compromising system performance, particularly in servers handling increasing workloads and data storage demands, as seen with rank multiplication which results in data bus contention and reduced streaming performance.

Innovation Solution

The implementation of rank aggregation, where multiple physical memory modules are combined to represent a larger logical rank, maintaining a one-to-one mapping between logical and physical ranks, allowing for expanded memory without the performance penalties associated with rank multiplication.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If rank multiplication is used to expand memory capacity, then memory capacity increases, but data bus contention occurs and streaming performance decreases

Engineering Contradiction:
Improvememory capacityVSAvoidstreaming performance
Core Design Contradiction:
Quantity of substanceVSProductivity

Solution Approach 1:

The patent segments the memory system into multiple independent memory channels, each handling a single rank. By dividing the memory controller into multiple channel controllers that independently manage separate memory modules, the system avoids data bus contention while maintaining high streaming performance. Each channel operates independently with its own memory interface, eliminating the performance degradation caused by rank multiplication.

Inventive Principle:
Principle #1Segmentation

2Quantity of substance

If memory capacity is expanded using traditional methods, then storage demand is met, but system performance is compromised

Engineering Contradiction:
Improvememory capacityVSAvoidsystem performance
Core Design Contradiction:
Quantity of substanceVSReliability

Solution Approach 1:

The patent transitions from expanding memory capacity within a single rank dimension to adding capacity across multiple channel dimensions. Instead of increasing the complexity of a single memory rank, the system adds parallel memory channels, each with its own controller and interface. This dimensional shift allows capacity expansion without compromising performance, as each channel operates independently without interfering with others.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

3Adaptability or versatility

If virtualization oversubscribes physical memory among multiple users, then memory utilization increases, but memory capacity requirements increase

Engineering Contradiction:
Improvememory utilizationVSAvoidmemory capacity requirements
Core Design Contradiction:
Adaptability or versatilityVSQuantity of substance

Solution Approach 1:

The patent creates a universal memory interface that can serve multiple virtual users through parallel channels. The memory system is designed to handle multiple memory requests simultaneously across different channels, providing multi-functionality that supports virtualization workloads. Each channel can be dynamically allocated to different virtual machines or users, allowing high memory utilization without requiring excessive physical memory capacity.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP2577480B1Memory expansion using rank aggregation
Publication Date: 2018.11.14 CISCO TECHNOLOGY INC
  • EP2577480B1 patent drawingFigure 1
  • EP2577480B1 patent drawingFigure 2
  • EP2577480B1 patent drawingFigure 3

AI summary

In one embodiment, a method includes receiving from a memory controller, a request to access memory stored at memory modules, the request directed to one of a plurality of logical ranks, mapping at a rank aggregator, the logical rank to one of a plurality of physical ranks at the memory modules, and forwarding the request to one of the memory modules according to the mapping. Two or more of the memory modules are combined to represent the number of logical ranks at the memory controller such that there is a one-to-one mapping between the logical ranks and the physical ranks. An apparatus for rank aggregation is also disclosed.