Storage Product with AI Accelerator and Segmented Message Paths
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional network-attached storage devices are inefficient due to the CPU being a bottleneck in processing power and communication bandwidth, especially when scaling up storage capacity, as they process both control and data messages, leading to high power consumption and increased workloads.
Innovation Solution
Implementing separate processing paths for control and data messages, where control messages are routed through the processing device for administrative and management operations, while data messages are directly communicated between the storage device and the storage client without going through the processing device, reducing the burden on the CPU and improving efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If the CPU processes both control and data messages in conventional network-attached storage devices, then administrative and management operations can be performed, but power consumption and processing workload increase significantly
Solution Approach 1:
The patent segments the message processing function into two independent paths: a control message processing path that routes through the CPU for administrative operations, and a data message processing path that bypasses the CPU. This segmentation allows the system to perform management operations while significantly reducing CPU involvement in data transfer operations, thereby lowering power consumption.
Solution Approach 2:
The patent extracts the data message processing function from the CPU processing path and creates a separate direct communication path between the network interface and storage device. This extraction removes the CPU as a bottleneck for data operations while preserving its role in control operations, reducing overall processing workload and power consumption.
2Ease of operation
If the CPU processes both control and data messages, then comprehensive control is achieved, but communication bandwidth requirements increase
Solution Approach 1:
The patent divides communication into two segmented paths: control messages that require CPU processing and data messages that use direct communication. This segmentation reduces the bandwidth burden on the CPU by eliminating data message routing through the processor, allowing comprehensive control capability with reduced communication bandwidth requirements.
3Quantity of substance
If storage capacity is scaled up in conventional systems, then more storage space is available, but the CPU bottleneck becomes more severe
Solution Approach 1:
The patent extracts data message processing from the CPU to enable scalable storage capacity without proportionally increasing processing requirements. By creating a direct communication path for data operations, the system can add storage capacity while maintaining constant processing efficiency, as the CPU is no longer the bottleneck for data transfer operations.
4Ease of operation
If the CPU processes all messages, then centralized control is maintained, but device complexity increases
Solution Approach 1:
The patent segments the processing architecture into two distinct paths with different complexity characteristics: a control path through the CPU for centralized management and a data path with direct communication for efficient transfers. This segmentation maintains centralized control capability while reducing overall device complexity by eliminating the need for CPU involvement in routine data operations.
Data Source
AI summary
A storage product manufactured as a standalone computer component and installed in a computing device having a computer bus connecting the storage product to a local host system. The storage product has an artificial intelligence accelerator, a network interface operable on a computer network, and a local storage device having a storage capacity accessible via the network interface and configured to store an artificial neural network model having instructions executable by the artificial intelligence accelerator. Over the computer bus the local host system controls access, made via the network interface, to the storage capacity. The storage product can perform, using the artificial intelligence accelerator, at least a portion of computations of the artificial neural network model to generate a neural network output from neural input data received via the network interface.