Distributed FPGA Cloud Integration via Signal Conversion

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In data centers, integrating Field Programmable Gate Arrays (FPGAs) is challenging due to the difficulty in accessing all available connections on data cards housed in rackmount servers, requiring significant modifications to existing hardware and limiting the seamless adoption of FPGAs in cloud computing environments.

Innovation Solution

A method and system that convert host interface connectors on data cards from one signaling standard to another, such as from PCI Express to Ethernet, allowing data cards to communicate and interconnect using different standards, and utilize a card management data processor to allocate functions and route data signals efficiently across multiple data cards, enabling flexible and high-speed data processing without substantial hardware modifications.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If data cards are housed in rackmount servers, then data center integration is achieved, but accessibility to connections becomes difficult

Engineering Contradiction:
Improvedata center integrationVSAvoidconnection accessibility
Core Design Contradiction:
Ease of manufactureVSEase of operation

Solution Approach 1:

The system segments the data card connections into two types: traditional host interface connectors for CPU communication and repurposed connectors for peer-to-peer data card communication. This segmentation allows each connector type to serve its specific function optimally while maintaining overall system integration.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Host interface connectors are given dual functionality: they continue to serve as interfaces between data cards and the host system while simultaneously enabling direct communication between data cards. This multi-functionality eliminates the need for separate connection interfaces for peer-to-peer communication.

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

2Productivity

If FPGA-powered appliances are increased, then computational capability improves, but hardware modifications become significant

Engineering Contradiction:
Improvecomputational capabilityVSAvoidhardware modifications
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

Data cards are equipped with self-sufficient capabilities including onboard memory, processing units, and direct communication interfaces. Each data card can independently perform data processing functions and communicate with other data cards without requiring constant host system intervention or complex external hardware modifications.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system employs dynamic function allocation where data cards can be reconfigured and reassigned to different processing tasks based on computational needs. The card management data processor dynamically manages resource allocation, allowing the system to adapt to varying computational demands without physical hardware changes.

Inventive Principle:
Principle #15Dynamics

3Device complexity

If signaling standards are standardized, then host communication simplifies, but interoperability with external devices becomes limited

Engineering Contradiction:
Improvehost communicationVSAvoidexternal device interoperability
Core Design Contradiction:
Device complexityVSAdaptability or versatility

Solution Approach 1:

The host interface connector serves as an intermediary that can translate between different signaling standards. It maintains standardized communication with the host system while simultaneously supporting multiple external signaling standards through the same physical interface, enabling versatile external device interoperability without compromising host communication simplicity.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentUS10834023B2Distributed FPGA solution for high-performance computing in the cloud
Publication Date: 2020.11.10 LDA TECH LTD
  • US10834023B2 patent drawing
  • US10834023B2 patent drawing
  • US10834023B2 patent drawing

AI summary

A data processing system, method and device. A device can include a plurality of data cards having host interface connectors initially configured to transmit signals according to a first communication protocol and data card connectors that communicate with external devices using a different communication protocol. The data cards are converted so that the host interface connectors also transmit signals using the second communication protocol. The plurality of data cards are interconnected so that signals can be routed through the data cards to provide desired data processing functions. A cross-point switch fabric allows the signals to be routed to the appropriate data card or cards. Multiple devices can be interconnected to provide a distributed data processing grid providing access to the data processing functions for external devices that do not communicate using the first communication protocol.