Storage Server Inference Engine Integration for Data Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current cloud computing systems incur high energy usage and latency due to the need to transfer large amounts of data from storage servers to remote inference servers for batch mode computations, which are not critical but still require significant resources.
Innovation Solution
Integrating inference engines directly into storage servers with byte-addressable memory, allowing for local decoding and processing of data without the need for remote transfer, thereby reducing data movement and energy consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If data is transferred from storage server to remote inference server for batch mode compute, then processing capability is improved, but energy consumption increases significantly (50-80% of overall energy usage)
Solution Approach 1:
The patent combines storage and compute functions into a single storage server by integrating inference engines directly into the storage server. This merging eliminates the need to transfer data between separate storage and inference servers, thereby reducing energy consumption associated with data movement while maintaining processing capability.
Solution Approach 2:
The patent introduces byte-addressable memory within the storage server, enabling compute operations to be performed directly on stored data without traditional data movement. This dimensional change in data access and processing architecture allows the system to maintain productivity while dramatically reducing energy loss.
2Productivity
If data is transferred from storage server to remote inference server, then compute tasks can be executed, but time latency increases due to data transfer requirements
Solution Approach 1:
By merging storage and compute capabilities into the same server, the patent eliminates data transfer latency between separate systems. The inference engine processes data directly where it is stored, removing the time penalty associated with network or bus transfers.
Solution Approach 2:
The byte-addressable memory structure is prepared in advance within the storage server, allowing compute operations to be initiated directly on stored data without preliminary data movement steps, thereby reducing overall task completion time.
3Productivity
If separate inference servers are deployed for batch mode compute, then processing power is increased, but system complexity increases
Solution Approach 1:
The patent merges storage and compute resources into a unified storage server architecture, eliminating the need for separate inference servers. This consolidation reduces system complexity while maintaining the processing power needed for batch mode compute tasks.
Solution Approach 2:
The storage server is designed with multi-functionality, serving both storage and compute roles. The inference engine integrated into the storage server enables the same hardware to perform both data storage and data processing functions, reducing the number of components needed in the system.
Data Source
AI summary
Examples herein relate to a solid state drive that includes a media, first circuitry, and second circuitry. In some examples, the first circuitry is to execute one or more commands. In some examples, the second circuitry is to receive a configuration of at one type of command, where the configuration is to define an amount of media bandwidth allocated for the at one type of command; receive a command; and assign the received command to the first circuitry for execution.