Dynamic Storage Accelerator Load Balancing via Latency Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In storage systems with multiple devices, accelerator resources are often unevenly utilized, leading to processing delays due to uneven job distribution, and existing job scheduling schemes fail to adaptively manage these resources effectively.
Innovation Solution
A method where a first device determines if its accelerator resource load exceeds a threshold, and if so, it sends a job processing request to a second device from a candidate list, receiving latency information to update the list, allowing for dynamic selection and improved resource utilization across devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If jobs are distributed across multiple devices, then device utilization is improved, but processing delay increases due to uneven distribution
Solution Approach 1:
The patent implements dynamic load balancing by continuously monitoring accelerator resource utilization across devices and adjusting job distribution in real-time. The system dynamically selects target devices based on current load conditions, transforming the static job distribution into an adaptive process that responds to changing system states, thereby maintaining both high utilization and low latency.
Solution Approach 2:
The system employs feedback mechanisms by monitoring accelerator resource utilization metrics and using this information to make informed decisions about job routing. The load balancing algorithm receives feedback about device performance and utilization levels, then adjusts subsequent job assignments to optimize overall system performance, resolving the contradiction between utilization and latency.
2Ease of operation
If accelerator resources are allocated to each device independently, then device autonomy is maintained, but resource utilization efficiency deteriorates due to uneven distribution
Solution Approach 1:
The patent creates a universal job scheduling system that can operate across multiple devices with independent accelerator resources. The load balancing mechanism serves multiple functions: it maintains device autonomy by allowing local execution, enables cross-device job migration when needed, and provides systematic resource allocation. This multi-functional approach allows the system to preserve device independence while achieving efficient resource utilization across the entire system.
Data Source
AI summary
Embodiments of the present disclosure relate to a method, an electronic device, and a computer program product for managing a storage system. The method includes: determining, at a first device of the storage system, whether a load of a first accelerator resource of the first device exceeds a load threshold; sending, if it is determined that the load exceeds the load threshold, a job processing request to a second device in a candidate device list to cause the second device to process a target job of the first device using a second accelerator resource of the second device, the candidate device list indicating devices in the storage system that can be used to assist the first device in job processing; receiving, from the second device, latency information related to remote processing latency of processing the target job using the second accelerator resource; and updating the candidate device list based on the latency information. The embodiments of the present disclosure can optimize the system performance.