Receiver-Based HPCC with Reduced NIC Feedback Overhead
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing congestion control algorithms in datacenter networks face challenges in achieving high throughput and ultra-low latency due to inherent limitations, including high packet-per-second overhead at receiver NICs and unfair bandwidth allocation among flows.
Innovation Solution
Implementing receiver-based High Precision Congestion Control (HPCC) that reduces feedback overhead by allowing receivers to determine transmit rates based on in-network telemetry, dynamically adjusting the additive-increase parameter to match current network conditions, and sending reduced feedback to senders.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional HPCC sends full operational and telemetry information feedback to senders, then measurement precision is improved, but packet-per-second overhead at receiver NICs increases
Solution Approach 1:
The patent extracts only the essential transmit rate control information from the full operational and telemetry data, sending selectively chosen parameters back to senders rather than all collected data. This reduces feedback overhead while maintaining sufficient precision for congestion control.
Solution Approach 2:
Instead of the traditional approach where senders determine transmit rates based on receiver feedback, this patent inverts the control logic by having receivers proactively send transmit rate directives to senders, reducing the need for extensive feedback packets and reversing the conventional congestion control information flow.
2Device complexity
If traditional congestion control uses fixed additive-increase parameter, then device complexity is reduced, but adaptability to changing network conditions deteriorates
Solution Approach 1:
The patent implements dynamic adjustment of the additive-increase parameter based on real-time network conditions observed through in-network telemetry. The parameter is modified adaptively rather than remaining fixed, allowing the system to respond to changing network states while maintaining manageable complexity through structured adjustment rules.
Solution Approach 2:
The patent changes the additive-increase parameter dynamically based on measured network conditions, adjusting this key control parameter to match current traffic patterns and congestion levels, thereby improving adaptability without requiring complete redesign of the congestion control mechanism.
3Measurement precision
If receiver NICs process high packet-per-second rates with full feedback, then measurement precision is maintained, but loss of time increases due to processing overhead
Solution Approach 1:
The patent extracts and sends back only the most critical transmit rate control information rather than complete operational data, reducing the processing time required at receiver NICs while preserving the essential feedback needed for accurate congestion control.
Solution Approach 2:
The patent applies partial feedback action by sending only the necessary portion of congestion control information back to senders, avoiding the excessive processing overhead of transmitting and analyzing complete operational and telemetry data while maintaining sufficient accuracy for effective congestion management.
Data Source
AI summary
Examples described herein relate to a network agent, when operational, to: receive a packet, determine transmit rate-related information for a sender network device based at least on operational and telemetry information accumulated in the received packet, and transmit the transmit rate-related information to the sender network device. In some examples, the network agent includes a network device coupled to a server, a server, or a network device. In some examples, the operational and telemetry information comprises: telemetry information generated by at least one network device in a path from the sender network device to the network agent.