CGRA Hot-Plug Resource Pooling for Multi-Tenant Runtime Continuity

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing cloud computing systems face challenges in dynamically managing reconfigurable processors due to hardware underutilization and inefficiencies in virtualization, particularly in multi-tenancy and dynamic workload scenarios, where practical virtualization support for accelerators has not kept pace with accelerator innovation, leading to hardware underutilization and performance gaps.

Innovation Solution

A technology that enables the dynamic removal and addition of Coarse-Grained Reconfigurable Array (CGRA) processors from a pool of such processors without shutting down the system, using a controller and runtime processor to manage hot-plug events, allocate resources, and ensure seamless operation through file descriptor data structures and interrupt handling.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If PCIe pass-through techniques are used to dedicate physical hardware to virtual machines, then hardware compatibility and isolation are improved, but hardware underutilization and loss of multi-tenancy occur

Engineering Contradiction:
Improvehardware isolationVSAvoidhardware utilization
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent segments reconfigurable processor resources into virtualized units that can be dynamically allocated to multiple virtual machines. Instead of dedicating entire physical hardware to single VMs, the system divides processor resources into shareable segments that maintain isolation through virtualization, enabling multi-tenancy while improving utilization.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements universal reconfigurable processor resources that can serve multiple functions and multiple virtual machines simultaneously. The same physical processor pool can be configured and allocated to different VMs based on workload requirements, providing multi-functionality and eliminating hardware underutilization while maintaining proper isolation through virtualization layers.

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Reliability

If reconfigurable processors are statically allocated to virtual machines, then hardware compatibility is improved, but adaptability to dynamic workloads deteriorates

Engineering Contradiction:
Improvehardware compatibilityVSAvoidworkload adaptability
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent implements dynamic resource allocation where reconfigurable processor resources can be allocated, deallocated, and reconfigured at runtime based on workload demands. The system allows VMs to request and receive processor resources dynamically without static pre-allocation, enabling adaptation to changing computational requirements while maintaining hardware compatibility through standardized virtualization interfaces.

Inventive Principle:
Principle #15Dynamics

3Adaptability or versatility

If practical virtualization support is added to keep pace with accelerator innovation, then workload adaptability is improved, but device complexity increases

Engineering Contradiction:
Improvevirtualization supportVSAvoidsystem complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent introduces a virtualization layer as an intermediary between physical reconfigurable processors and virtual machines. This intermediary layer handles resource management, allocation, and configuration tasks, shielding the complexity of accelerator hardware from VMs while providing standardized access interfaces. The virtualization layer acts as a mediator that simplifies the overall system architecture despite the underlying hardware complexity.

Inventive Principle:
Principle #24Intermediary (Mediator)

4Productivity

If reconfigurable processors are removed or added during operation, then resource utilization is improved, but system stability may deteriorate

Engineering Contradiction:
Improveresource utilizationVSAvoidsystem stability
Core Design Contradiction:
ProductivityVSStability of the object's composition

Solution Approach 1:

The patent implements hot-plug capability that allows reconfigurable processors to be added or removed from the system during operation without shutting down. The virtualization layer prepares and manages resource allocation in advance, ensuring that dynamic resource changes occur smoothly with proper handling of configuration loading and task migration, thereby maintaining system stability during resource transitions.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentUS12511174B2Hot-plug events in a pool of reconfigurable data flow resources
Publication Date: 2025.12.30 SAMBANOVA SYSTEMS INC
  • US12511174B2 patent drawing
  • US12511174B2 patent drawing
  • US12511174B2 patent drawing

AI summary

A data processing system comprises a pool of reconfigurable data flow resources with arrays of physical configurable units, a controller, and a runtime processor. The controller is configured to generate a hot-plug event in response to detecting a removal of an unallocated array of physical configurable units from the pool of reconfigurable data flow resources. The runtime processor is configured to execute user applications on a subset of the arrays of physical configurable units and to receive the hot-plug event from the controller. The runtime processor is further configured to make the removed unallocated array of physical configurable units unavailable for subsequent allocations of subsequent virtual data flow resources and subsequent executions of subsequent user applications, while the subset of the arrays of physical configurable units continues the execution of the user applications.