Memory Controller Throttling for Balanced Workload Distribution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing memory systems face challenges in efficiently managing and processing requests from multiple submission queues to prevent intensive processing of host commands, which can lead to imbalanced workload distribution and increased processing loads.
Innovation Solution
The implementation of a memory system with a controller that acquires requests from submission queues, generates commands for nonvolatile memory, and controls throttling based on the number of commands and requests in the system, ensuring balanced processing among queues.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the memory system acquires requests from multiple submission queues without throttling control, then the processing speed and productivity are improved, but the workload distribution becomes imbalanced and the processing load on the controller increases excessively
Solution Approach 1:
The controller monitors the number of requests in each submission queue and adjusts the acquisition rate dynamically. When a queue has fewer requests than a threshold, the controller reduces acquisition from that queue to prevent excessive workload concentration, thereby maintaining balanced workload distribution while preserving overall processing efficiency
Solution Approach 2:
The request acquisition mechanism transitions from a static uniform approach to a dynamic adaptive approach. The controller continuously adjusts the acquisition rate for each submission queue based on real-time request counts, enabling the system to maintain optimal productivity while adapting to varying workload conditions across different queues
2Ease of operation
If the controller processes all requests from all submission queues uniformly, then the ease of operation is maintained, but the processing load on the controller becomes excessive and workload distribution becomes imbalanced
Solution Approach 1:
The uniform processing control is segmented into queue-specific acquisition control. Instead of applying a single uniform acquisition rate to all submission queues, the controller divides the control mechanism to allow independent adjustment of acquisition rates for each queue based on its current request count, thereby reducing overall processing load while maintaining operational simplicity
Data Source
AI summary
According to one embodiment, a memory system includes a nonvolatile memory, a storage area, and a controller. The controller acquires a request from a submission queue included in a host, generates one or more commands to be executed by the nonvolatile memory in accordance with the request, and stores the commands to the storage area. The controller controls throttling of acquisition of requests from the submission queue in accordance with the number of commands in the storage area and the number of requests in the submission queue.