PCIe Fabric Storage Access for High-Bandwidth Compute Nodes

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing network storage architectures struggle to provide sufficient bandwidth for high-demand applications like cloud gaming, particularly with the increasing complexity of online gaming services, as gigabit Ethernet connections fall short of the required performance.

Innovation Solution

Implementing a PCI Express (PCIe) fabric within a rack assembly to enable direct access between compute nodes and network storage, achieving greater than 4 Gigabyte per second (GB/s) network storage bandwidth per compute node at non-volatile memory express (NVMe) latencies.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If gigabit Ethernet connections are used for network storage, then network connectivity is provided, but bandwidth performance is insufficient for high-demand applications

Engineering Contradiction:
Improvebandwidth performanceVSAvoidnetwork architecture complexity
Core Design Contradiction:
SpeedVSDevice complexity

Solution Approach 1:

The patent introduces a PCIe fabric as an intermediary communication medium between compute nodes and network storage, replacing traditional Ethernet switching. This fabric enables direct-attached storage performance over the network by establishing dedicated PCIe pathways, thereby achieving high bandwidth without increasing overall system complexity through standardized interface adoption.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent changes the communication protocol parameter from Ethernet TCP/IP to PCIe protocol, enabling transmission speeds exceeding 4 GB/s per compute node. This parameter change transforms the network storage interface capabilities while maintaining compatibility through standardized PCIe implementations across different hardware components.

Inventive Principle:
Principle #35Parameter changes

2Productivity

If faster bandwidth performance is implemented, then cloud gaming capabilities are enhanced, but network connection requirements become more demanding

Engineering Contradiction:
Improvecloud gaming performanceVSAvoidnetwork connection demand
Core Design Contradiction:
ProductivityVSUse of energy by moving object

Solution Approach 1:

The patent segments the network storage access into dedicated PCIe pathways for each compute node, allowing parallel data transfers that enhance cloud gaming performance. This segmentation enables multiple compute nodes to simultaneously access network storage at high speeds without contending for shared bandwidth, thereby improving productivity while managing energy demand through efficient resource allocation.

Inventive Principle:
Principle #1Segmentation

3Speed

If PCI Express fabric is implemented for high-speed access, then storage bandwidth exceeds 4 GB/s per compute node, but implementation complexity increases

Engineering Contradiction:
Improvestorage access speedVSAvoidimplementation complexity
Core Design Contradiction:
SpeedVSDevice complexity

Solution Approach 1:

The patent implements a universal PCIe fabric architecture that serves multiple functions: high-speed network storage access, inter-node communication, and device I/O. This multi-functional design achieves storage access speeds exceeding 4 GB/s per compute node while reducing implementation complexity by using a single standardized PCIe interface throughout the system rather than multiple specialized interfaces.

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

Data Source

PatentEP4104395B1Network architecture providing high speed storage access through a PCI express fabric between a compute node and a storage server within an array of compute nodes
Publication Date: 2025.12.03 SONY INTERACTIVE ENTERTAINMENT LLC
  • EP4104395B1 patent drawingFigure 1
  • EP4104395B1 patent drawingFigure 2A
  • EP4104395B1 patent drawingFigure 2B

AI summary

A network architecture including a streaming array that includes a plurality of compute sleds, wherein each compute sled includes one or more compute nodes. The network architecture including a network storage of the streaming array. The network architecture including a PCIe fabric of the streaming array configured to provide direct access to the network storage from a plurality of compute nodes of the streaming array. The PCIe fabric including one or more array-level PCIe switches, wherein each array-level PCIe switch is communicatively coupled to corresponding compute nodes of corresponding compute sleds and communicatively coupled to the network storage. The network storage is shared by the plurality of compute nodes of the streaming array.