Coprocessor Schedule Controller for Interleaved Workload Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Coprocessors face performance issues when handling multiple calculation-intensive workloads simultaneously due to sequential processing and the need for entire workloads to be loaded into shared memory before processing can begin, leading to delays and resource management inefficiencies.
Innovation Solution
A schedule controller manages context storage within a coprocessor's first storage unit to interleave workloads, allowing for simultaneous processing by segmenting workloads into commands and utilizing a second storage unit to free up space when the first unit is full, enabling immediate processing of additional workloads.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the coprocessor processes workloads sequentially to completion, then processing reliability is maintained, but productivity deteriorates when multiple calculation-intensive workloads need to be run simultaneously
Solution Approach 1:
The patent segments workloads into smaller command units that can be independently managed and processed. The command queue divides workload data into discrete commands, allowing the coprocessor to process multiple workloads in an interleaved manner rather than completing one entire workload before starting the next, thus improving productivity while maintaining processing reliability through structured command management
Solution Approach 2:
The patent implements dynamic workload management through a schedule controller that dynamically assigns commands from multiple workloads to the coprocessor engine based on availability and priority. This dynamic scheduling allows the system to adaptively switch between different workload commands, enabling simultaneous processing of multiple calculation-intensive workloads while maintaining system reliability
2Reliability
If the entire workload is loaded into shared memory prior to processing, then processing reliability is ensured, but loss of time increases due to delays in loading large workloads
Solution Approach 1:
The patent applies preliminary action by pre-loading only the necessary command units into the command queue before processing begins, rather than requiring the entire workload to be loaded into shared memory first. This allows the coprocessor to start processing available commands immediately while the rest of the workload data is still being transferred, significantly reducing the time loss associated with workload loading
Solution Approach 2:
By segmenting the workload into smaller command units that can be independently queued and processed, the system loads and processes commands incrementally rather than requiring complete workload loading. This segmentation enables the coprocessor to begin processing as soon as the first commands are available, maintaining processing reliability through structured command execution while eliminating the time penalty of loading entire large workloads before processing
3Manufacturing precision
If the coprocessor waits for the entire workload to be available in shared memory, then processing accuracy is maintained, but productivity deteriorates due to idle waiting time
Solution Approach 1:
The system performs preliminary actions by preparing and queuing command units in advance based on available data, allowing the coprocessor to start processing immediately when commands are ready rather than waiting for complete workload availability. This maintains processing accuracy through proper command sequencing while improving productivity by eliminating idle waiting time
Solution Approach 2:
The dynamic command queue management system continuously monitors data availability and dynamically adjusts the processing schedule, enabling the coprocessor to process available command units as they become ready. This dynamic approach maintains processing accuracy through structured command execution while maximizing productivity by keeping the coprocessor continuously engaged with available work rather than idle
Data Source
AI summary
Methods and apparatus of interleaving two or more workloads are presented herein. The methods and apparatus may comprise a schedule controller and a coprocessor. The schedule controller is operative to utilize the first storage unit to manage context stored therein that allows for the coprocessor to interleave the two or more workloads that can be directly supported by the first storage unit. The coprocessor includes a dedicated first storage unit and an engine.