Trace Buffer Event Timing for Multi-Calculator Fault Isolation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In computer systems with multiple calculation units, identifying faults and grasping the internal state is difficult due to independent data movement without a host unit, making it challenging to locate faults and understand data flow.
Innovation Solution
A computer system with trace buffers in each calculation unit records detection times of events based on operating frequency, allowing the host unit or calculation units to manage and erase trace data under specific conditions, facilitating fault identification and data flow management.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If calculation units independently move data without host unit involvement, then data processing speed is improved, but fault identification capability deteriorates
Solution Approach 1:
A trace buffer is introduced as an intermediary component in each calculation unit to record detection times of events. This trace buffer acts as a mediator that captures processing information locally, enabling fault identification without requiring continuous host unit involvement in data movement operations.
Solution Approach 2:
The trace buffer records detection times of events in advance before faults occur. By preliminarily capturing processing information including timestamps and event types, the system can later identify faults by analyzing recorded data without needing to interrupt ongoing independent data movement operations.
2Difficulty of detecting and measuring
If trace data is retained for fault analysis, then fault identification capability is improved, but memory usage and power consumption increase
Solution Approach 1:
The system selectively discards trace data that is no longer needed for fault analysis. The trace buffer manages retention periods for detection times, discarding old or redundant data while preserving critical information needed for identifying faults, thus balancing memory usage and power consumption against fault identification capability.
3Measurement precision
If multiple detection times are recorded for each input data type, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The trace buffer records multiple detection times as copies for each input data type, creating a historical record of when events were detected. This copying approach preserves precise timing information for multiple events without requiring complex real-time analysis, as the trace buffer simply stores sequential detection time copies that can be analyzed later.
Data Source
AI summary
An embodiment is a method for controlling a computer system including a plurality of calculators and a host, each of the plurality of calculators having a trace buffer, the method including triggering by detection of a predetermined event from input data to record a detection time of the event acquired for each type of the input data in the trace buffer, determining, by the host or the calculator, the type of the input data to which the detection time of the event recorded in the trace buffer belongs, and recording, by the host or the calculator, according to a determination result, a latest detection time of the event among the detection times of the event belonging to the same type of the input data, and erases detection times of the event other than the latest detection time of the event.