Direct-Mapped Flash Storage for Hyperscale AI Infrastructure

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Traditional storage systems face inefficiencies in data management and storage operations, particularly in handling flash drives without integrated storage controllers, leading to redundant processes and reduced reliability.

Innovation Solution

Implementing a direct-mapped flash storage system where the operating system initiates and controls processes, such as data rewriting and erasure, without address translation by storage controllers, and utilizing non-volatile RAM as a quickly accessible buffer to improve write latency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Extent of automation

If storage controllers perform address translation and control flash drives, then data management is automated, but system complexity increases and reliability decreases due to redundant processes

Engineering Contradiction:
Improvedata management automationVSAvoidsystem complexity
Core Design Contradiction:
Extent of automationVSDevice complexity

Solution Approach 1:

The patent extracts the storage controller functionality from the flash drive, allowing the operating system to directly manage flash drives without an intermediate storage controller layer. This eliminates redundant address translation processes and reduces system complexity while maintaining automation through OS-level control.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The operating system assumes multiple roles including address translation and flash drive management that were previously handled by dedicated storage controllers. This consolidation allows the OS to directly control flash drives, reducing the number of components and simplifying the overall system architecture.

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

2Ease of operation

If storage controllers manage data rewriting and erasure processes, then flash drive operations are controlled, but write latency increases due to additional processing steps

Engineering Contradiction:
Improveflash drive controlVSAvoidwrite latency
Core Design Contradiction:
Ease of operationVSLoss of time

Solution Approach 1:

The patent removes the storage controller intermediary that caused write latency. By allowing the operating system to directly initiate and control data rewriting and erasure processes on flash drives, the system eliminates the additional processing steps and address translation overhead that previously increased write latency.

Inventive Principle:
Principle #2Taking out (Extraction)

3Adaptability or versatility

If traditional storage controllers are used, then flash drives can be managed, but reliability decreases due to redundant write operations

Engineering Contradiction:
Improveflash drive compatibilityVSAvoidflash drive reliability
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The patent extracts the storage controller layer that introduced reliability issues through redundant write operations. By enabling direct OS control over flash drives, the system eliminates the intermediary that caused unnecessary writes, thereby improving reliability while maintaining flash drive compatibility through standardised OS-level interfaces.

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentUS11494692B1Hyperscale artificial intelligence and machine learning infrastructure
Publication Date: 2022.11.08 PURE STORAGE INC
  • US11494692B1 patent drawing
  • US11494692B1 patent drawing
  • US11494692B1 patent drawing

AI summary

A hyperscale artificial intelligence and machine learning infrastructure includes a plurality of racks, where: at least one or more of the racks include one or more GPU servers; at least one or more of the racks include one or more storage systems; each of the racks include one or more switches coupled to at least one switch in another rack; and the one or more GPU servers are configured to execute one or more artificial intelligence or machine learning applications, wherein data stored within the one or more storage systems is used as input to the one or more artificial intelligence or machine learning applications.