Communication Bypass Apparatus for High-Performance Computing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In high-performance computing systems, optical communication links often fail due to high failure rates of optical modules, leading to link deactivation and communication disruptions, which are difficult to recover from without redundant lanes or significant performance degradation.
Innovation Solution
An inter-node communication device and method that performs communication path switching and resource reallocation by detecting failed paths, bypassing errors, and reallocating resources to maintain communication between nodes, even when all lanes in a link fail, using a service processor and test circuit to manage error detection and path switching.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If optical communication is used to achieve large bandwidth, then communication bandwidth is improved, but reliability deteriorates due to high failure rate of optical modules
Solution Approach 1:
The link is divided into multiple independent lanes (e.g., 4 lanes per link). When failure occurs in one lane, only that specific lane is deactivated while other lanes remain operational, preventing complete link failure and maintaining partial communication capacity
Solution Approach 2:
The system performs communication tests and monitors error counters for each lane in advance. When error rates exceed thresholds, the system proactively deactivates problematic lanes before complete failure occurs, preventing link degradation and maintaining reliability
2Reliability
If dynamic lane degeneracy is performed to handle lane failures, then reliability is improved by maintaining communication, but productivity deteriorates due to loss of communication capacity
Solution Approach 1:
The system dynamically adjusts the number of active lanes based on real-time error conditions. Error counters are continuously monitored, and lanes are selectively deactivated only when needed, allowing the system to maintain maximum communication capacity under normal conditions while adapting to failures when they occur
Solution Approach 2:
Instead of deactivating entire links upon lane failure, the system performs partial deactivation of only the specific failed lanes. This partial action maintains communication capacity through remaining lanes, minimizing productivity loss while ensuring reliability
3Reliability
If link deactivation is performed when all lanes fail, then reliability is improved by preventing error propagation, but productivity deteriorates due to complete loss of communication path
Solution Approach 1:
The interconnect control unit implements universal error handling across all links and lanes through standardized error counters and threshold-based deactivation. This unified approach ensures consistent reliability protection while minimizing productivity impact by activating bypass paths through other links when available
Data Source
AI summary
An information processing apparatus includes a first and a second node devices, and a control device configured to control data transmission between the first and the second node devices, the control device being coupled to the first node device through a first path group including a plurality of paths and being coupled to the second node device through a second path group including a plurality of paths, the control device is configured to perform a communication test of the first path group and the second path group, and when a first failure is detected in a first path in the first path group, couple a third path other than the first path in the first path group with the first node device, couple a second path in the second path group with the second node device, and couple the third path and the second path with each other.