Multi-Root PCIe Switch via Network Tunneling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional I/O connection systems are limited in connecting and distributing multiple computers and I/Os on a large scale due to physical restrictions and the need for modified software and hardware to accommodate longer-range connections, which is not feasible with existing PCIe standards.
Innovation Solution
A multi-root PCI Express (PCIe) switch is implemented using a network with distributed elements, allowing multiple computers and I/Os to share a single I/O while maintaining PCIe standards compliance, avoiding PCIe bus timeouts, and eliminating the need for hardware and software modifications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional PCIe connection systems are used to connect multiple computers and I/Os, then the connection can be established within a single chassis, but the system is physically restricted and cannot scale to large-scale distributed connections
Solution Approach 1:
The system segments the traditional monolithic PCIe switch into multiple distributed PCIe switch elements that can be geographically separated and connected via network infrastructure. Each element maintains independent PCIe switching functionality while participating in a unified logical switch through network-based interconnection, enabling large-scale distributed connections without physical chassis limitations.
Solution Approach 2:
A network intermediary layer is introduced between distributed PCIe switch elements to enable communication and coordination. This intermediary translates and routes PCIe transactions across network infrastructure, allowing elements to function as a unified switch while being physically distributed, thereby extending connection distance without compromising adaptability.
2Adaptability or versatility
If the circuit scale of the switch is increased to accommodate more computers and I/Os, then the connection capacity increases, but the physical size and complexity of the switch device increases proportionally
Solution Approach 1:
The switch functionality is segmented into multiple smaller PCIe switch elements distributed across different physical locations. Each element handles a subset of computers and I/Os, avoiding the need for a single large complex switch. The network interconnection provides the additional capacity needed for large-scale connections without increasing individual device complexity.
Solution Approach 2:
The system transitions from a two-dimensional planar switch architecture to a three-dimensional distributed architecture by adding the network communication dimension. This allows connection capacity to scale by distributing elements spatially and utilizing network infrastructure, rather than increasing the circuit scale within a single device plane.
3Length of stationary object
If PCIe bus timeout issues occur during TLP transmission over long distances, then connection reliability deteriorates, but extending connection distance is necessary for large-scale deployment
Solution Approach 1:
Network intermediary elements are positioned between distributed PCIe switch elements to manage TLP transmission. These intermediaries handle timeout management, TLP encapsulation/decapsulation, and coordination of transactions across network boundaries, ensuring reliable communication over extended distances without PCIe bus timeout issues.
Solution Approach 2:
The patent replaces the direct electrical PCIe bus connection with network-based communication for long-distance TLP transmission. This substitution eliminates the electrical signal degradation and timeout issues inherent in direct PCIe connections by using network protocols designed for longer-distance communication, while maintaining PCIe compatibility through encapsulation.
Data Source
AI summary
Upstream network interfaces (2-1-2-N) and downstream network interfaces (5-1-5-M) have an upstream PCI-PCI bridge function and a downstream PCI-PCI bridge function, respectively. These network interfaces (2-1-2-N, 5-1-5-M) and a network (3) are incorporated in a system as a single multi-root PCI express switch. The network (3) tunnels TLPs (Transaction Layer Packets) between the upstream network interfaces (2-1-2-N) and the downstream network interfaces (5-1-5-M) or between the downstream network interfaces (5-1-5-M). This enables to distribute and connect a plurality of computers and a plurality of I/Os on a large scale without changing software, root complexes, and I/Os.