Computational Storage Request Batching to Reduce Fabric Bottlenecks

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The handling of data near storage devices is inefficient when near-storage processing is used, leading to fabric bottlenecks and redundant data transmission due to multiple individual requests and responses.

Innovation Solution

A mechanism to build batch requests that combine multiple individual requests, allowing the computational storage unit to execute them serially, in parallel, or in a hybrid manner, reducing redundant data transmission and optimizing fabric usage.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If multiple individual requests are sent to the storage device, then the storage device can process each request independently, but the fabric load and bandwidth usage increase due to redundant data transmission

Engineering Contradiction:

Improverequest processing independenceVSAvoidfabric bandwidth usage

Core Design Contradiction:

ReliabilityVSLoss of energy

Solution Approach 1:

The patent combines multiple individual requests into a single batch request that can be processed by the storage device. The batch request includes multiple sub-requests (read, write, compute operations) that are aggregated and transmitted together over the fabric, eliminating redundant data transmission while maintaining independent processing of each sub-request by the storage device's computational engine.

Inventive Principle:

Principle #5Merging (Combining)

2Ease of operation

If multiple individual requests are sent to the storage device, then each request can be executed separately, but the overall processing time increases due to multiple round trips

Engineering Contradiction:

Improverequest execution flexibilityVSAvoidoverall processing time

Core Design Contradiction:

Ease of operationVSLoss of time

Solution Approach 1:

The system performs preliminary actions by aggregating multiple requests into a batch request before transmission. The host processor prepares the batch request containing multiple sub-requests and transmits it in a single operation, eliminating the need for multiple separate round trips between the host and storage device, thereby reducing overall processing time while maintaining execution flexibility.

Inventive Principle:

Principle #10Preliminary action

3Productivity

If near-storage processing is used for individual requests, then data processing can be performed close to the storage medium, but fabric bottlenecks occur due to repeated data transmission

Engineering Contradiction:

Improvenear-storage processing capabilityVSAvoidfabric bottleneck

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent merges multiple near-storage processing requests into a single batch request that leverages the storage device's computational engine. By processing multiple operations (read, write, compute) in a batch manner within the storage device itself, the system utilizes near-storage processing capabilities efficiently without causing fabric bottlenecks, as data remains localized and is not repeatedly transmitted over the fabric.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS12443436B2Systems and method for batching requests in computational devices

Publication Date: 2025.10.14 SAMSUNG ELECTRONICS CO LTD

AI summary

A system is described. The system may include a processor. Software may run on the processor to build a batch request, which may include at least a first request and a second request. The system may also include a storage device and a computational storage unit associated with the storage device, which may be configured to execute the batch request.