Snoop Control Section Error Recovery in Multiprocessor Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional multiprocessor systems face challenges in recovering from synchronization errors during broadcast processing, as errors within global ports cannot be effectively recovered from, leading to unprocessable memory access requests due to exclusive access control or resource contention.
Innovation Solution
The system control apparatus and method implement a mechanism where the snoop control section detects synchronization errors, outputs error retry instructions to local ports, and resets global ports, allowing retry processing to commence from the local port of the request source, thereby increasing the possibility of error recovery.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If snooping is performed synchronously in all SCs with memory access requests being broadcast to global ports, then snooping control is facilitated, but synchronization errors cannot be recovered from when they occur in global ports
Solution Approach 1:
The patent segments the broadcast processing into two distinct parts: (1) broadcast output from local ports to global ports, and (2) snooping execution based on requests from global ports. This segmentation allows error recovery to occur at the local port level without affecting the entire snooping system, thereby resolving the contradiction between ease of control and error recovery capability
Solution Approach 2:
The patent implements preliminary error detection by comparing request IDs before executing snooping operations. By detecting potential synchronization errors in advance and preventing erroneous snooping execution, the system maintains reliability while preserving the simplicity of synchronous control
2Productivity
If memory access requests are broadcast to all SCs through global ports, then snooping can be performed across all cache memories, but synchronization errors lead to unprocessable requests and system halting
Solution Approach 1:
The patent implements feedback mechanisms where SCs broadcast their request IDs to all other SCs, and each SC compares received request IDs with its own. This feedback loop enables detection of synchronization errors and triggers appropriate error handling, allowing the system to maintain productivity while ensuring reliability through continuous monitoring and correction
Solution Approach 2:
The patent prepares error handling procedures in advance by establishing request ID comparison and error detection protocols before snooping operations begin. This beforehand cushioning ensures that when synchronization errors occur, the system can immediately activate recovery procedures rather than halting, thus maintaining system continuity
Data Source
AI summary
A system control apparatus and method capable of increasing the possibility of recovery from a synchronization error in snooping between system controllers are provided. The system control apparatus has a local port that holds a memory access request received externally and reoutputs it in response to an error retry instruction. A broadcast output section broadcasts the memory access request. A broadcast input section receives the broadcast memory access request. A global port holds the received memory access request. A snoop control section transmits and receives cache status information containing a result of snooping, and detects a synchronization error on the basis of the cache status information. If a synchronization error occurs, the snoop control section outputs the error retry instruction to the local port holding the memory access request that resulted in the synchronization error.