Dynamic Resource Isolation for Data Processing Accelerators
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data processing accelerators, such as AI accelerators, lack effective isolation mechanisms, allowing malicious entities to access and exploit AI models and data through communication channels, compromising security.
Innovation Solution
Implementing dynamic resource isolation and static partitioning of resources within data processing accelerators, using virtual functions and a resource management unit to manage and secure access, ensuring that virtual machines can only access assigned resources and dynamically adjusting resource allocations as needed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If communication channels are enabled between DPAs for sensitive transactions, then data processing capability is improved, but security and isolation between DPAs deteriorates
Solution Approach 1:
The DPA is segmented into multiple isolated partitions, each handling specific transactions. The segmentation includes separating AI model storage, data processing units, and communication interfaces into distinct security zones that cannot be accessed by other partitions, thus maintaining security while enabling communication channels.
Solution Approach 2:
A security management unit acts as an intermediary between different DPA partitions and external interfaces. This intermediary controls and monitors all communication channels, enforcing access policies and preventing unauthorized access to AI models and data while allowing legitimate data processing operations.
2Reliability
If resource isolation mechanisms are implemented in DPAs, then security is improved, but device complexity increases
Solution Approach 1:
The security management unit performs multiple functions including access control, encryption key management, transaction monitoring, and partition isolation enforcement. By consolidating these security functions into a single multi-functional unit, the design achieves strong isolation without proportionally increasing overall device complexity.
Solution Approach 2:
The isolation mechanism uses a nested hierarchical structure where security contexts are nested within partitions, which are nested within the DPA. Each level provides a layer of isolation, with the security management unit nesting at the core to manage all security policies, creating efficient layered protection without redundant complexity.
Data Source
AI summary
Embodiments of the disclosure discloses a method and system of a virtualization environment for a data processing (DP) accelerator. In one embodiment, a data processing (DP) accelerator includes a resource management unit and one or more dynamically isolated resources managed by the resource management unit. The DP accelerator includes one or more virtual functions (VFs) each associated with one of the one or more dynamically isolated resources, where a virtual machine (VM) of a host is assigned one of the one or more VFs to access the dynamically isolated resources associated with the assigned VF, and where the VM has no access to the rest of the one or more dynamically isolated resources.


