Collaboration Server Switch Fabric Interconnect
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current cluster node servers for video conferencing suffer from poor performance per Watt and performance per rack unit, and multiprocessor servers require expensive cache-coherent CPUs, leading to high costs and inefficiencies.
Innovation Solution
A collaboration server design featuring a switch fabric interconnecting multiple processing modules, each with a CPU core, GPU core, and MPU core, eliminating the need for cache coherence and enabling dynamic task assignment and load balancing across modules, using a tree topology of PCIe switches for efficient data transmission and processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Stability of the object's composition
If multiprocessor servers use cache-coherent CPUs to enable shared memory space, then system integration and memory sharing are improved, but CPU cost and power consumption increase significantly
Solution Approach 1:
The system divides memory into isolated memory spaces, each associated with a processing module, eliminating the need for cache coherence between modules. This segmentation allows independent memory management without the overhead of maintaining cache coherence across the entire system, reducing power consumption while maintaining functional integrity.
Solution Approach 2:
The switch fabric acts as an intermediary between processing modules, enabling communication and data exchange without requiring shared memory or cache coherence. This mediator approach allows modules to interact efficiently through the switch fabric while maintaining isolated memory spaces, avoiding the power overhead of cache-coherent architectures.
2Stability of the object's composition
If multiprocessor servers use cache-coherent CPUs, then memory sharing is enabled, but device cost increases due to specialized CPUs
Solution Approach 1:
The system segments memory into isolated spaces per processing module, eliminating the need for expensive cache-coherent CPUs. Standard CPUs can be used in each module since they manage their own isolated memory spaces, significantly reducing hardware costs while maintaining system functionality.
Solution Approach 2:
Instead of using expensive specialized cache-coherent CPUs, the system uses multiple standard CPUs, each managing its own isolated memory space. This copying approach replicates the memory management functionality across modules without requiring expensive specialized hardware, reducing overall system cost.
3Adaptability or versatility
If cluster node servers use traditional architecture, then video conferencing functionality is provided, but performance per rack unit deteriorates
Solution Approach 1:
The server is divided into multiple independent processing modules, each capable of handling video conferencing tasks. This segmentation increases density and performance per rack unit while maintaining full video conferencing functionality through coordinated operation of modules via the switch fabric.
Solution Approach 2:
The system transitions from a traditional monolithic server architecture to a modular architecture organized in groups across multiple racks. This dimensional reorganization allows better utilization of space and resources, improving performance per rack unit while maintaining video conferencing capabilities.
4Adaptability or versatility
If cluster node servers use traditional architecture, then video conferencing is supported, but performance per Watt deteriorates
Solution Approach 1:
The system segments processing into multiple independent modules with isolated memory spaces, eliminating the power overhead of cache coherence mechanisms. Each module uses standard CPUs that consume less power, while the switch fabric enables efficient inter-module communication, improving performance per Watt for video conferencing workloads.
Solution Approach 2:
The switch fabric serves as an energy-efficient intermediary for inter-module communication, replacing the power-intensive cache coherence protocols of traditional multiprocessor systems. This mediator approach maintains video conferencing functionality while significantly reducing power consumption.
Data Source
Figure 1
Figure 2
AI summary
A collaboration server comprising: a switch fabric; and a plurality of processing modules, each processing module including a CPU core; wherein the plurality of processing modules are interconnected via the switch fabric.