Hardware Resource Sharing via Remote Orchestrator Configuration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies lack the ability to share processing or I/O hardware resources between nodes or allocate I/O hardware resources to a pool of available resources, limiting fine-level partitioning and resource sharing in data centers.
Innovation Solution
A device capable of hardware-based inter-device resource sharing, utilizing a remote orchestrator to configure hardware resources for local or remote operation, enabling sub-socket level partitioning and resource pooling through interface technologies like PCIe, allowing nodes to share processing and I/O resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If software virtualization is used for resource sharing, then flexibility of data processing is improved, but hardware functionality is abstracted out and made unavailable
Solution Approach 1:
The patent segments hardware resources into distinct pools (compute resources, storage resources, I/O resources) that can be independently managed and shared. Each resource type is virtualized separately through its own configuration space, allowing selective sharing while maintaining direct hardware access paths when needed, thus preserving hardware functionality while enabling flexibility.
Solution Approach 2:
The patent introduces configuration spaces as intermediary structures between hardware resources and processing nodes. These configuration spaces act as mediators that enable resource sharing while maintaining direct hardware access paths, allowing hardware functionality to remain available while supporting virtualized access when required.
2Ease of operation
If coarse granularity partitioning is used, then platform level resource access is simplified, but fine-level partitioning capability is lost
Solution Approach 1:
The patent implements hierarchical segmentation of resource partitioning, allowing both coarse-grain platform-level partitioning and fine-grain sub-socket level partitioning within the same system. Each level operates independently through its own configuration space, enabling simultaneous use of both partitioning granularities without interference.
Solution Approach 2:
The patent adds a new dimension to resource partitioning by introducing configuration spaces that operate at multiple hierarchical levels. This enables the system to partition resources both at the platform level (coarse granularity) and at the socket/sub-socket level (fine granularity), effectively adding a temporal or hierarchical dimension to the partitioning capability.
3Productivity
If hardware resources are dedicated to a particular node, then processing performance is improved, but resource sharing between nodes is limited
Solution Approach 1:
The patent implements dynamic resource allocation where hardware resources can transition between dedicated and shared states based on configuration space settings. Processing nodes can dynamically access resources from other nodes through the interconnect fabric when needed, allowing the system to adapt resource allocation to workload requirements while maintaining high performance through direct access paths.
Solution Approach 2:
The patent makes hardware resources universal by enabling them to serve multiple functions: dedicated local processing, shared access by other nodes, and pooling into resource pools. The configuration spaces enable the same physical resource to be accessed differently by different nodes, providing multi-functionality without sacrificing processing performance.
4Adaptability or versatility
If physical reconfiguration is performed for resource sharing, then resource allocation flexibility is improved, but system downtime and complexity increase
Solution Approach 1:
The patent replaces physical reconfiguration mechanisms with software-based configuration space management. Instead of mechanically moving or physically reconfiguring hardware resources, the system uses programmable configuration spaces to logically allocate and share resources, dramatically reducing complexity and enabling dynamic reconfiguration without physical intervention.
Solution Approach 2:
The patent performs resource allocation and sharing configuration in advance through configuration spaces before actual resource usage. This preliminary configuration establishes access paths and resource mappings that remain valid during operation, eliminating the need for complex real-time physical reconfiguration and reducing system downtime.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present disclosure is directed to hardware-based inter-device resource sharing. For example, a remote orchestrator (RO) may provide instructions to cause a device to make at least one hardware resource available to other devices. An RO module in the device may interact with the RO and may configure a configuration module in the device based on instructions received from the RO. The configuration module may set a device configuration when the device transitions from a power off state to a power on state. The device may also comprise a processing module to process data based on the device configuration, interface technology (IT) and at least one hardware resource. The interface technology may allow the processing module and the at least one hardware resource to interact. The RO module may configure the IT to allow the at least one hardware resource to operate locally or remotely based on the instructions.