Cloud Accelerator Manager Trace Analysis for Failure Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Cloud-based hardware accelerators can cause failures in computer programs running in the cloud, and existing technologies lack effective methods to detect and diagnose these failures efficiently.
Innovation Solution
A cloud-based accelerator manager that monitors computer programs, generates traces of accelerator usage, and analyzes these traces to determine if failures are due to accelerator issues, using OpenCAPI interfaces to dynamically generate and deploy accelerators, and manage their usage within the cloud environment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If cloud-based hardware accelerators are used to improve run-time performance, then processing speed is improved, but system reliability deteriorates due to undetected accelerator failures
Solution Approach 1:
The patent implements a feedback mechanism where the accelerator manager continuously monitors accelerator usage and generates traces of accelerator operations. When a program failure is detected, the system retrieves and analyzes the corresponding accelerator trace to determine if the failure originated from the accelerator, creating a closed-loop feedback system that improves reliability while maintaining performance benefits
Solution Approach 2:
The accelerator manager acts as an intermediary component between the cloud-based accelerators and the host system. It introduces a trace buffer and analysis mechanism that mediates between accelerator operations and program execution, allowing the system to maintain high-speed accelerator processing while adding reliability through trace-based failure detection without directly modifying the accelerator hardware
2Reliability
If accelerator usage is monitored and traced to detect failures, then system reliability is improved, but device complexity increases
Solution Approach 1:
The patent extracts the monitoring and trace analysis functionality into a separate accelerator manager component, isolating the complexity from the core accelerator and program execution paths. The trace buffer and analysis logic are taken out as independent modules that can be added or removed without affecting the fundamental accelerator operation, reducing the perceived complexity in the main system
Solution Approach 2:
The system creates copies of accelerator usage information in the form of traces stored in a trace buffer. Instead of modifying the accelerator hardware to include monitoring capabilities, the system copies the necessary operational data to analysis structures that can be examined for failure detection, maintaining simplicity in the original accelerator while adding reliability through data duplication
Data Source
AI summary
A cloud-based accelerator manager manages cloud-based hardware accelerators. The accelerator manager monitors computer programs running in the cloud and generates a trace that indicates which accelerators were used and when. The trace may be for a single computer program or may be for multiple computer programs. Thus, the trace can be program-specific, showing all accesses to accelerators by a single program, or may be accelerator-specific, showing all accesses to each accelerator by all computer programs. The cloud-based accelerator manager detects a failure in one of the computer programs executing in the cloud. The cloud-based accelerator manager provides the trace to a user, who can then analyze the trace to determine whether the failure was due to an accelerator. The cloud-based accelerator manager thus helps detect when cloud-based accelerators are the reason for failures for computer programs running in a cloud.