Splitter Device Asymmetric Data Stream Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional information handling systems face inefficiencies in processing network data streams due to latency and bandwidth costs associated with copying and communicating data between processors via interconnects, as well as the need for each processor to obtain its own copy of the data stream.
Innovation Solution
A multi-processor/endpoint data splitting system that uses a splitter device within a multi-endpoint adapter to asymmetrically or symmetrically split incoming network data streams, directing each portion directly to its respective processing subsystem without the need for copying across processors, utilizing Direct Memory Access (DMA) operations and configuration notifications to optimize data distribution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If network data streams are copied between processors via UPIs for separate processing, then multiple processors can process different portions of the data stream, but latency and bandwidth costs increase
Solution Approach 1:
The network data stream is segmented into different portions (header/metadata portions and payload portions) that are assigned to different processors for parallel processing. The UPI interconnect enables this segmentation by allowing each processor to access only its designated data portions, eliminating the need to copy the entire data stream across all processors.
Solution Approach 2:
Each processor is given local access to specific portions of the network data stream through the UPI interconnect, rather than requiring all processors to have access to all data. This local quality approach reduces bandwidth consumption and latency by ensuring each processor only processes its designated data portions.
2Productivity
If network data streams are copied between processors via UPIs for separate processing, then multiple processors can process different portions of the data stream, but bandwidth costs increase
Solution Approach 1:
The network data stream is segmented into different portions (header/metadata portions and payload portions) that are assigned to different processors for parallel processing. The UPI interconnect enables this segmentation by allowing each processor to access only its designated data portions, eliminating the need to copy the entire data stream across all processors.
Solution Approach 2:
Each processor is given local access to specific portions of the network data stream through the UPI interconnect, rather than requiring all processors to have access to all data. This local quality approach reduces bandwidth consumption and latency by ensuring each processor only processes its designated data portions.
3Ease of operation
If each processor obtains its own copy of the network data stream via the network, then processors can independently process data, but inefficiencies and increased network traffic occur
Solution Approach 1:
The UPI interconnect acts as an intermediary between the network interface controller and the processors, enabling efficient data distribution without requiring each processor to obtain its own copy via the network. The UPI provides a direct, high-speed pathway that eliminates redundant network traffic while maintaining independent processor operation.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A multi-endpoint adapter device (202a) includes a splitter device (322) that is coupled to a network port and a plurality of endpoint subsystems (320a, 320c) that are each coupled to a processing subsystem (303). The splitter device (322) receives, via the network port, a first data payload (508), and identifies both a first data sub-payload (510a) that is included in the first data payload (508) and that is associated with a first endpoint subsystem (320a) included in the plurality of endpoint subsystems and a second data sub-payload (510b) that is included in the first data payload (508) and that is associated with a second endpoint subsystem (320c) included in the plurality of endpoint subsystems. The splitter device (322) then splits the first data payload (508) into the first data sub-payload (510a) and the second data sub-payload (510b), and forwards both the first data sub-payload (510a) to the first endpoint subsystem (320a) and the second data sub-payload (510b) to the second endpoint subsystem (320c).