NVMe-oF GPU Data Processing for CDN Latency

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional content delivery networks (CDNs) face inefficiencies in content distribution due to excessive data processing steps, leading to high energy consumption and costs, as content data often passes through multiple touchpoints before reaching the end user.

Innovation Solution

Implementing a nonvolatile memory express over fabrics (NVMe-oF) system with embedded or chassis-integrated graphics processing units (GPUs) that process and transfer data directly to end users, bypassing local CPUs and reducing unnecessary data movements.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Use of energy by moving object

If content data passes through multiple touchpoints (CPU, memory, network interface) before reaching the end user, then data processing and transfer can be performed, but energy consumption and network latency increase

Engineering Contradiction:

Improveenergy consumptionVSAvoidnetwork latency

Core Design Contradiction:

Use of energy by moving objectVSLoss of time

Solution Approach 1:

The patent extracts the data processing function from the CPU and places it directly in the storage device. The storage device now performs data processing operations (such as compression, encryption, or format conversion) internally before transferring data, eliminating the need for data to travel to and from the CPU and reducing both energy consumption and latency.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent merges the data processing function with the storage device by integrating a processing unit within the storage device. This combination allows data to be processed and transferred in a single operation without requiring separate CPU involvement, thereby reducing the number of touchpoints and improving efficiency.

Inventive Principle:

Principle #5Merging (Combining)

2Ease of manufacture

If standard servers and processors are used to perform content distribution tasks, then content can be delivered, but costs and energy consumption increase

Engineering Contradiction:

Improvecost effectivenessVSAvoidenergy consumption

Core Design Contradiction:

Ease of manufactureVSUse of energy by moving object

Solution Approach 1:

The patent makes the storage device multi-functional by enabling it to perform both data storage and data processing operations. This eliminates the need for separate dedicated servers or processors, reducing hardware costs and energy consumption while maintaining the ability to deliver content effectively.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The storage device serves itself by performing data processing operations internally without requiring external CPU assistance. This self-service capability reduces dependency on expensive and energy-intensive standard servers, lowering both operational costs and energy consumption.

Inventive Principle:

Principle #25Self-service

Data Source

PatentUS11997163B2System and method for supporting energy and time efficient content distribution and delivery

Publication Date: 2024.05.28 SAMSUNG ELECTRONICS CO LTD

AI summary

A method of transferring data to an end user via a content distribution network using an nonvolatile memory express over fabrics (NVMe-oF) device, the method including receiving a read request at the NVMe-oF device, translating a logical address corresponding to the data to a physical address, fetching the data from a flash storage of the NVMe-oF device, processing the data with a GPU that is either embedded in the NVMe-oF device, or on a same chassis as the NVMe-oF device, and transferring the data.