Top-of-Rack Switch Flow Pinning for Collective Traffic Balancing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for transmitting flows of communication collectives in networks often result in network congestion due to uneven distribution of flows, leading to poor transport performance and tail-latency issues, especially in leaf-spine topologies, without requiring additional NIC support or incurring undue cost.
Innovation Solution
A network switch determines flow groups with common sources and destinations, pinning each flow to a corresponding link to evenly distribute them across links, ensuring balanced traffic and preventing congestion by leveraging the correlation of flows within collectives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If flows are distributed across shared links using conventional hashing methods, then flow distribution is achieved, but network congestion occurs and transport performance deteriorates
Solution Approach 1:
The system performs preliminary identification of collective communication flows and their associated topologies before routing decisions are made. By detecting flows that belong to the same collective and identifying their temporal and destination correlations in advance, the system can proactively distribute them across different links to prevent congestion before it occurs, rather than reacting to congestion after it has formed.
Solution Approach 2:
The system segments the network traffic by identifying and grouping flows that belong to the same collective communication operation. By separating these correlated flows into distinct groups based on their temporal and destination correlations, the system can apply different routing strategies to different groups, distributing them across multiple links to avoid the congestion that occurs when all flows are treated uniformly.
2Device complexity
If flows are spread across multiple links without NIC hardware support, then hardware cost is reduced, but packets may arrive out of order reducing performance
Solution Approach 1:
The system performs preliminary identification and grouping of flows that belong to the same collective before they are transmitted. By detecting the temporal and destination correlations of flows in advance and grouping them together, the system ensures that correlated packets are transmitted together and arrive together at the destination, eliminating the out-of-order delivery problem without requiring special NIC hardware support for packet reordering.
3Ease of operation
If only first-hop link congestion information is used for flow assignment, then routing decisions are simplified, but subsequent link congestion cannot be accounted for
Solution Approach 1:
The system performs preliminary identification of collective communication flows and their topologies before routing decisions are made. By detecting flows that belong to the same collective and identifying their temporal and destination correlations in advance, the system can proactively distribute them across different links to prevent congestion before it occurs, rather than reacting to congestion after it has formed.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A network device such as a top-of-rack (TOR) switch receives communication flows for communication collectives from source devices and determines, by importing from the source devices or snooping from the communication traffic, topology information of the communication flows. For each communication collective, the TOR switch determines, based at least in part on the topology information, groups of communication flows that are correlated in time and in destination. For each respective group, the TOR switch pins each communication flow in the respective group to a corresponding network link connected to the TOR switch so that the communication flows in the respective group are evenly distributed across network links connected to the TOR switch.