CPU-GPU Lockstep Controller with Backup Allocation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In lockstep computing systems, when multiple processors are run in parallel, if one processor fails, the redundancy fails, leading to loss of error correction and redundancy, making it difficult to detect and correct errors efficiently.
Innovation Solution
A processing system with three or more CPU-GPU pairs and a backup CPU, where a lockstep controller operates the pairs in parallel, compares outputs, and upon detecting a defective pair, discontinues its operation, reallocates resources by assigning GPU address spaces to ensure continued operation with the backup CPU, maintaining system reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple processors are run in parallel in a lockstep system, then error detection capability is improved, but system reliability deteriorates when one processor fails
Solution Approach 1:
The patent implements a backup processor that is pre-configured and ready to take over immediately upon detecting a failure in the active processors. This preliminary preparation ensures that when a processor fails, the system can quickly switch to the backup without loss of redundancy or error detection capability, thus resolving the contradiction between improved error detection and maintained reliability.
Solution Approach 2:
The system incorporates a backup processor as a cushion against processor failures. This backup unit is maintained in a ready state, providing a safety margin that compensates for the potential failure of any single active processor. The beforehand cushioning ensures that the lockstep system maintains its reliability and error detection capabilities even when one of the parallel processors fails.
2Reliability
If a backup processor is added to maintain redundancy, then system reliability is improved, but device complexity increases
Solution Approach 1:
The patent merges the backup processor with the existing lockstep system architecture, integrating it into the same computational framework. The backup processor shares the same hardware resources, memory space, and control mechanisms as the active processors, allowing it to function as a seamless extension of the system. This merging approach improves reliability while minimizing the increase in device complexity by avoiding separate, isolated backup infrastructure.
Solution Approach 2:
The backup processor is designed with universal functionality, capable of performing the same tasks as any active processor. It can assume the role of any failed processor in the lockstep configuration, making it a multi-functional component that provides redundancy without requiring specialized hardware. This universality reduces the complexity burden by using the same architectural building blocks for both active and backup roles.
3Measurement precision
If real-time failure detection is implemented, then error detection speed is improved, but processing time is increased
Solution Approach 1:
The lockstep system continuously executes the same operations on multiple processors simultaneously, maintaining continuous computation rather than intermittent checking. This continuous execution enables real-time failure detection through output comparison while ensuring that the system is always productive. The useful action of computing continues uninterrupted, and failure detection is performed as part of the normal operational flow, thus improving detection speed without significant processing time penalty.
Solution Approach 2:
The system implements feedback mechanisms where the outputs of parallel processors are continuously compared to detect failures. This feedback loop operates in real-time during the execution of computational tasks, allowing immediate detection of processor failures. The feedback is integrated into the lockstep architecture itself, using the computational outputs as the feedback signal, which minimizes additional processing time while maintaining fast error detection.
Data Source
AI summary
A lockstep controller operates a lockstep system of three or more CPU-GPU pairs, comparing the outputs from the CPU-GPU pairs and, by way of a majority vote, provides the output for the lockstep system. Based on comparing the outputs, if one of the CPU-GPU pairs provides outputs that disagree with the majority outputs, it can be switched out of the lockstep system. The removed CPU is replaced by a backup CPU. So that the backup CPU can be part of a CPU-GPU pair, a portion of the address space from the GPU of one of the other CPU-GPU pairs is assigned to the backup CPU to operate as a replacement CPU-GPU pair, while the CPU already associated with this GPU retains another portion of the GPU's address space to continue operating as a CPU-GPU pair.