Multi-Stage Network Flow Scheduling Using Edge-Based Delay Probing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network control systems face limitations in controlling packet transmissions when network components are closed environments, and clock synchronization inaccuracies lead to jitter and biased processing in multi-stage interconnection networks, particularly in high-intensity applications like large language models, where switches are not individually accessible for synchronization.
Innovation Solution
Implement edge-based flow scheduling using a netcam module that monitors and controls data traffic at the application layer, detecting path utilization and performing scheduling without relying on in-network support, and utilizing a clock synchronization system for precise timestamping across hosts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing control systems operate at the control layer to control packet transmissions, then network control functions can be performed, but controllers cannot access closed environments such as NICs where instructions travel directly from memory to the NIC with no intermediate point for intervention
Solution Approach 1:
The patent moves the control point from the traditional control layer (layer 3) to the application layer (layer 7), creating a new dimensional approach to network control. By implementing control logic at the application layer, the system can intercept and control network traffic before it enters closed environments like NICs, bypassing the accessibility limitations of lower layers.
Solution Approach 2:
The patent introduces an intermediary control mechanism at the application layer that acts as a mediator between the application and the network stack. This intermediary can observe and control network traffic without requiring direct access to closed network components, effectively bridging the gap between control needs and accessibility constraints.
2Measurement precision
If clock synchronization is used to determine transaction order in finance and e-commerce, then transaction sequencing can be established, but clock synchronization inaccuracies result in jitter and biased or non-optimal processing
Solution Approach 1:
The patent applies preliminary scheduling actions at the application layer before traffic enters the network. By determining transaction order and applying scheduling decisions in advance, the system eliminates the need for reliance on imprecise clock synchronization during network transmission, thereby reducing jitter and processing delays.
Solution Approach 2:
The patent creates a logical copy of the network traffic flow at the application layer, allowing control and scheduling decisions to be made on this copy without affecting the actual network transmission timing. This enables precise transaction ordering to be established independently of physical clock synchronization accuracy.
3Productivity
If in-network support from switches is used to perform OWD calculations, then flow scheduling can be performed, but actions are taken without central knowledge and switches must be individually accessible
Solution Approach 1:
The patent merges the flow scheduling functionality from distributed switches into a centralized controller at the application layer. By consolidating scheduling decisions in one central location with global knowledge of network state, the system eliminates the need for individual switch accessibility and enables coordinated control across the entire network.
Solution Approach 2:
The patent extracts the flow scheduling logic from the network infrastructure (switches) and relocates it to the application layer. This extraction removes the dependency on in-network support and individual switch accessibility, allowing flow scheduling to be performed centrally based on comprehensive network knowledge.
Data Source
AI summary
For each respective flow over a network initiated by an application layer, a controller determines a respective path taken by the respective flow from respective source host to respective destination host, transmits respective probe packets along the respective path while taking timestamps at the respective source host and the respective destination host, and determines a respective one-way delay for the respective path based on the timestamps. The controller determines utilization for each path, determines determining an optimal usage across each path, and schedules transmission of data for each respective flow based on the optimal usage.