I/O Routing in Multidimensional Torus Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Massively parallel computing systems face inefficiencies in routing I/O packets between compute nodes and I/O nodes, as current systems require separate networks for point-point, collective, and I/O communications, leading to increased costs and complexity.
Innovation Solution
A method is introduced to route data packets in a multidimensional torus network by assigning a toio value to each packet, allowing it to be routed to an associated I/O node, and using ioreturn values to route packets back through the torus network, with certain torus links on I/O nodes configured as additional I/O links, enabling each I/O node to be attached to multiple compute nodes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If separate networks are used for point-point, collective, and I/O communications, then communication reliability is improved, but device complexity and cost increase
Solution Approach 1:
The patent combines point-point, collective, and I/O communication networks into a single unified torus network. Compute nodes and I/O nodes share the same network infrastructure, eliminating the need for separate physical networks. This merging reduces device complexity and cost while maintaining communication reliability through unified network management and routing protocols.
Solution Approach 2:
The torus network is designed to perform multiple communication functions simultaneously - point-point messaging, collective operations, and I/O communications - all over the same network infrastructure. The network achieves multi-functionality through configurable routing and node roles, allowing a single network to replace multiple specialized networks.
2Adaptability or versatility
If packets are re-injected into the network for I/O communication, then routing flexibility is improved, but memory and processor utilization increase
Solution Approach 1:
The patent pre-configures I/O links directly at compute nodes, establishing dedicated communication paths to I/O nodes before packets need to be transmitted. By pre-establishing these links and configuring routing information in advance, the system avoids the need to re-inject packets into the network, thereby reducing memory and processor utilization while maintaining routing flexibility for I/O communications.
3Adaptability or versatility
If each I/O node is attached to multiple compute nodes, then network versatility is improved, but device complexity increases
Solution Approach 1:
I/O nodes are designed with multi-functionality to handle communications with multiple compute nodes simultaneously. Each I/O node can serve multiple compute nodes through the shared torus network, achieving network versatility without requiring dedicated physical connections to each compute node. This universal design reduces device complexity compared to one-to-one mappings.
Data Source
AI summary
A method, system and computer program product are disclosed for routing data packet in a computing system comprising a multidimensional torus compute node network including a multitude of compute nodes, and an I/O node network including a plurality of I/O nodes. In one embodiment, the method comprises assigning to each of the data packets a destination address identifying one of the compute nodes; providing each of the data packets with a toio value; routing the data packets through the compute node network to the destination addresses of the data packets; and when each of the data packets reaches the destination address assigned to said each data packet, routing said each data packet to one of the I/O nodes if the toio value of said each data packet is a specified value. In one embodiment, each of the data packets is also provided with an ioreturn value used to route the data packets through the compute node network.