Virtualized DPU Management OS Recovery Across Mixed DPU Hardware

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The integration of data processing units (DPUs) from various manufacturers into virtualization and management solutions is challenging due to their diverse interfaces and operations, leading to difficulties in fully utilizing their capabilities within host devices.

Innovation Solution

A DPU management operating system virtual machine is launched using a DPU management hypervisor executed as firmware or privileged microcode, enabling advanced lifecycle management and recovery, which includes a DPU management operating system that communicates with a management hypervisor to provide access to physical resources and manage workloads across different DPU devices.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If DPUs from various manufacturers are integrated into virtualization solutions, then device versatility is improved, but device complexity increases due to diverse interfaces and operations

Engineering Contradiction:
ImproveDPU compatibilityVSAvoidmanagement complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

A management hypervisor is introduced as an intermediary layer between the management service and diverse DPU devices. This hypervisor abstracts manufacturer-specific interfaces and operations, providing a unified management interface while handling the complexity of communicating with different DPU types underneath. The hypervisor translates standardized management commands into manufacturer-specific operations, resolving the contradiction between supporting multiple DPU vendors and maintaining simple management.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If DPU management operating system updates are performed, then system reliability is improved, but productivity decreases due to downtime

Engineering Contradiction:
Improvesystem stabilityVSAvoidoperational continuity
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The management hypervisor performs preliminary actions by preparing and validating DPU management operating system updates before applying them to production systems. Updates are staged and tested in advance, and the hypervisor coordinates update deployment to minimize disruption. This preliminary preparation ensures system reliability improvements while reducing the actual downtime during update application.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The management hypervisor maintains continuous management operations during DPU updates by orchestrating sequential updates across multiple DPUs rather than updating all simultaneously. This allows the system to maintain operational continuity with partial DPU availability while still achieving reliability improvements through updated management operating systems.

Inventive Principle:
Principle #20Continuity of useful action

Data Source

PatentUS20250342050A1Lifecycle and recovery for virtualized DPU management operating systems
Publication Date: 2025.11.06 VMWARE INC
  • US20250342050A1 patent drawing
  • US20250342050A1 patent drawing
  • US20250342050A1 patent drawing

AI summary

Disclosed are various examples of lifecycle and recovery management for virtualized data processing unit (DPU) management operating systems. A DPU device executes a DPU management hypervisor that communicates with a management service over a network. The DPU management hypervisor virtualizes DPU hardware resources and passes control of the virtualized DPU hardware resources to a DPU management operating system (OS) virtual machine (VM). The DPU management hypervisor maintains control of a management network interface card (NIC) of the DPU device.