Shared Memory Error Recovery for Multiprocessor Partitions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multiprocessor systems face challenges in error recovery, particularly in isolating faults within partitions to allow unaffected partitions to continue using shared resources without disrupting system throughput, and require extensive logic and dedicated interfaces for error reporting and recovery.
Innovation Solution
A system and method for error recovery in a multiprocessor system where shared memory units process requests from multiple partitions, allowing non-affected partitions to continue using resources by isolating the effects of a fault through shared queue and state logic management, using existing interfaces and logic to handle errors without dedicated interfaces, and employing maintenance processors for diagnostic and recovery operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the common resource discontinues processing requests from both failing and non-failing partitions after a fault is detected, then the fault effects are isolated from shared resources, but system throughput is impacted due to temporary stoppage of non-failing partitions
Solution Approach 1:
The patent segments the error recovery process into two independent phases: (1) removing fault effects from shared resources, and (2) re-initializing the failing partition's logic. This allows non-failing partitions to continue processing during phase (1), eliminating the need to stop all partitions for fault isolation.
Solution Approach 2:
The patent introduces a maintenance processor as an intermediary component that handles the re-initialization of the failing partition's logic. This mediator can operate independently of the failing partition and shared resources, enabling recovery without disrupting non-failing partitions.
2Reliability
If dedicated error reporting and recovery interfaces are provided for partition-by-partition error recovery, then error recovery can be performed on a partition basis, but power utilization increases and logic resources are consumed
Solution Approach 1:
The patent makes the existing request/response interfaces universal by enabling them to serve dual purposes: normal data processing and error recovery. The same interfaces used for routine operations are reused for fault notification and recovery, eliminating the need for dedicated error recovery interfaces.
Solution Approach 2:
The patent merges error recovery functionality into the existing request/response processing logic. By combining normal operation and error recovery handling in the same logic paths, the system reduces the overall logic resources required while maintaining partition-based recovery capability.
Data Source
AI summary
A system and method is disclosed for performing error recovery in a data processing system that supports multiple processing partitions. One or more processors and I/O modules, as well as a portion of the address space of a main memory, is allocated to each partition. In this type of configuration, requests generated by units of multiple partitions are processed by the same queue and state logic of the main memory. When a failure occurs within one processing partition, one or more units are identified as being directly affected by the fault. All requests and responses from, and to, the affected units, as well as any logical residue of these requests and responses are removed from the shared memory queue and state logic in a manner that allows the other partition to continue issuing requests and responses to the memory in a normal manner that does not involve recovery operations.