Dual-DPU Storage Control for I/O Stream Processing Bottlenecks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The current storage system architecture centered on CPU computing struggles to efficiently process I/O data streams due to stagnated computing capabilities, leading to bottlenecks in performance improvement.
Innovation Solution
Implementing a storage control node with a first and second Data Processing Unit (DPU) to handle I/O requests and data processing, respectively, where the first DPU performs protocol offloading and data message storage, and the second DPU performs data processing, utilizing high-speed interconnection buses and built-in acceleration engines to optimize data paths.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a CPU-centered architecture is used for storage system, then control and management functions can be centralized, but data stream processing efficiency deteriorates due to stagnated CPU computing capability
Solution Approach 1:
The patent divides the storage control node into multiple functional units: CPU for control management, first DPU for protocol offloading, and second DPU for data processing. This segmentation allows each component to specialize in specific tasks, improving data stream processing efficiency while maintaining centralized control through the CPU.
Solution Approach 2:
The patent introduces DPUs as intermediary components between the CPU and storage devices. The first DPU acts as an intermediary for protocol offloading, and the second DPU serves as an intermediary for data processing operations. These intermediaries handle data-intensive tasks, freeing the CPU from bottlenecks while maintaining system coordination.
2Productivity
If all computation and control are initiated by the CPU, then system management is simplified, but I/O data stream processing performance deteriorates due to CPU bandwidth limitations
Solution Approach 1:
The patent extracts computation-intensive tasks from the CPU and assigns them to dedicated DPUs. The first DPU extracts protocol offloading functions, and the second DPU extracts data processing functions. This extraction reduces CPU computational load and energy consumption while improving I/O data stream processing performance.
Solution Approach 2:
The DPUs are designed to autonomously handle their respective tasks without requiring constant CPU intervention. The first DPU independently performs protocol offloading, and the second DPU independently executes data processing operations, enabling self-service functionality that reduces overall system overhead and improves performance.
3Speed
If protocol offloading and data processing are performed by separate units, then processing parallelism is improved, but device complexity increases
Solution Approach 1:
The patent segments processing functions into two specialized DPUs: the first DPU for protocol offloading and the second DPU for data processing. This segmentation enables parallel execution of these functions, improving processing speed while keeping the number of additional units minimal (only two DPUs added to the existing CPU-Storage architecture).
Data Source
AI summary
A data stream processing method, a storage control node, and a nonvolatile readable storage medium are provided. The storage control node includes a first Data Processing Unit (DPU) and a second DPU. The method includes: the first DPU receives an Input/Output interface (I/O) request, performs protocol offloading on the I/O request to obtain a data message, and stores the data message in a memory; and the second DPU acquires the data message from the memory and performs data processing on the acquired data message.