Latency-Aware Fabric Routing for Consistent Low-Latency Paths
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Disaggregated scheduled fabric (DSF) topology experiences latency inconsistencies due to varied cable lengths and different paths, which are detrimental to latency-sensitive workloads.
Innovation Solution
Implement a leaf node with a routing logic that classifies network paths into low, medium, and high latency groups based on latency attributes, using a routing table to select optimal paths for traffic routing, ensuring consistent low-latency connectivity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If cables of fixed or varied length are used to connect spine switches and leaf switches in DSF topology, then network scalability and flexibility are improved, but latency inconsistencies occur due to different path lengths
Solution Approach 1:
The routing logic pre-calculates and classifies network paths into latency groups before traffic arrives. By performing this classification in advance and storing it in a routing table, the system prepares optimal path selections beforehand, eliminating the need for real-time latency calculations and ensuring consistent low-latency routing for time-sensitive workloads
Solution Approach 2:
The routing logic segments network paths into distinct latency groups (e.g., low-latency group, medium-latency group, high-latency group) based on measured latency attributes. This segmentation allows the system to select appropriate path groups based on workload requirements, isolating latency-sensitive traffic from paths with higher variability
2Reliability
If multiple network paths are available between leaf switches, then network redundancy and load balancing capability are improved, but path selection complexity increases due to latency variations across paths
Solution Approach 1:
The routing logic performs preliminary classification of all available network paths into latency-based groups and stores this information in a routing table. When traffic arrives, the system simply looks up the appropriate pre-classified path group based on workload type, avoiding complex real-time calculations and reducing the computational burden of path selection
Solution Approach 2:
The routing logic changes the routing parameter from dynamic real-time latency calculation to static pre-classified latency groups. By measuring latency attributes once and categorizing paths into discrete groups, the system simplifies the routing decision process while maintaining awareness of latency characteristics across multiple redundant paths
Data Source
AI summary
Existing network architectures may experience latency variabilities due to varied cable lengths and diverse network paths available in a network fabric. Therefore, devices, systems, methods, and processes for facilitating enhanced latency-aware routing are described herein. A routing logic in a network fabric classifies two or more network paths, coupling a first network device to a second network device, into a plurality of groups based on latency attributes of the network paths. The routing logic in response to receiving a traffic flow at the first network device for the second network device, selects from the plurality of groups, a target group and routes the traffic flow to the second network device via at least one network path in the target group. If the target group includes more than one network path, the routing logic executes load balancing among a set of network paths classified into the target group.