On-Chip Interconnect Control for Multicast Write Port Bottlenecks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In on-chip interconnection networks, data transmission efficiency is reduced due to the need for destination devices to use a unified write port, leading to lower utilization of write ports and increased waiting times during multicasting or broadcasting.
Innovation Solution
A data transmission system where a processor configures transmission control information to source and destination processing elements, allowing the source to determine data packets and transmit them through an interconnection network, and the destination to store data based on separate control information, without requiring a unified write port for the destination devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If destination devices use a unified write port for data transmission, then data transmission can be performed, but write port use efficiency is reduced and transmission efficiency is reduced
Solution Approach 1:
The patent segments the write port resource allocation by introducing separate control information for each destination device. Instead of forcing all destinations to share a single write port, the system divides the control mechanism so that each destination device can be assigned its own write port through individualized control information from the source device, thereby eliminating the bottleneck of unified write port usage.
Solution Approach 2:
The patent implements dynamic write port assignment where the source device can flexibly select and assign different write ports to different destination devices based on real-time port availability. This dynamic allocation allows the system to adapt to changing conditions and maximize write port utilization, rather than being constrained by a static unified write port requirement.
2Productivity
If destination devices are constrained to use write ports with the same identifier, then data transmission can be performed, but use efficiency of write ports is reduced
Solution Approach 1:
The patent segments the control mechanism by providing separate control information for each destination device. This allows the source device to independently specify different write port identifiers for different destinations, breaking the constraint of unified write port usage and enabling each destination to utilize appropriate write ports based on their availability and requirements.
Solution Approach 2:
The patent applies local quality by allowing different write port assignments for different destination devices. Each destination device can be assigned a write port that is locally optimal for its specific situation, rather than being forced to use a globally unified write port. This localized optimization improves overall system adaptability and write port utilization.
Data Source
AI summary
Embodiments of this disclosure disclose a data transmission system and method, a medium, a device, and a chip. An apparatus includes: a processor; and at least two processing elements, which are interconnected to form an interconnection network. The processor is configured to configure first transmission control information to a first processing element that serves as a source device, and configure second transmission control information to a second processing element that serves as a destination device. The first processing element is configured to: when serving as the source device, determine a to-be-transmitted first data packet based on the first transmission control information, and transmit the first data packet to the second processing element through the interconnection network. The second processing element is configured to: when serving as the destination device, store at least a part of data in the first data packet based on the second transmission control information.


