Partition-Aware Broadcast Filtering for Cache Slice Isolation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In multiprocessor systems, broadcast operations such as cache state management and memory synchronization can interfere between partitions, leading to performance degradation and security vulnerabilities, especially in virtualized environments where malicious processes can exploit these operations for denial-of-service attacks.

Innovation Solution

A method is implemented to determine a partition identifier for broadcast operations, generate a sub-partition mask based on this identifier, and transmit the operation only to the subset of system components assigned to the user process, thereby isolating and filtering broadcast operations on a per-operation basis.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If broadcast operations are performed to manage cache state and synchronize memory across the entire memory subsystem, then memory synchronization and coherence are improved, but partition isolation deteriorates and interference between partitions increases

Engineering Contradiction:
Improvememory synchronizationVSAvoidpartition interference
Core Design Contradiction:
ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The patent segments the memory subsystem into partition-specific subsets based on the issuing partition's allocated memory regions. Instead of broadcasting to the entire memory subsystem, the broadcast operation is divided and transmitted only to the specific subset of memory components assigned to the issuing partition, thereby maintaining synchronization while eliminating unnecessary interference with other partitions.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by tailoring the broadcast operation scope to the specific local memory regions associated with the issuing partition. The broadcast is customized to affect only the local memory subset relevant to the issuing partition, rather than uniformly affecting the entire memory subsystem, thus reducing harmful interference while preserving necessary coherence.

Inventive Principle:
Principle #3Local quality

2Stability of the object's composition

If broadcast operations are transmitted to the entire memory subsystem, then memory coherence is maintained, but performance deteriorates due to unnecessary processing in other partitions

Engineering Contradiction:
Improvememory coherenceVSAvoidprocessing efficiency
Core Design Contradiction:
Stability of the object's compositionVSProductivity

Solution Approach 1:

The patent segments the broadcast operation into partition-specific portions by determining the issuing partition's allocated memory regions and transmitting the broadcast only to those specific regions. This segmentation eliminates unnecessary processing in other partitions while preserving coherence for the relevant memory regions, thereby improving overall processing efficiency.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies partial action by transmitting the broadcast operation only to the extent necessary for the issuing partition's allocated memory regions, rather than broadcasting to the entire memory subsystem. This partial broadcast reduces unnecessary processing overhead in other partitions while maintaining sufficient coherence for the affected regions.

Inventive Principle:
Principle #16Partial or excessive action

3Object-affected harmful factors

If partition isolation is enhanced by filtering broadcast operations, then security and interference reduction are improved, but device complexity increases

Engineering Contradiction:
Improvepartition isolationVSAvoidbroadcast filtering mechanism
Core Design Contradiction:
Object-affected harmful factorsVSDevice complexity

Solution Approach 1:

The patent implements universality by using a unified broadcast filtering mechanism that operates across all memory subsystem components. The filtering logic is integrated into the existing broadcast infrastructure, allowing the same mechanism to handle different partition configurations and memory layouts without requiring separate specialized systems for each partition isolation scenario.

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

Solution Approach 2:

The patent applies preliminary action by pre-determining the issuing partition's allocated memory regions before executing the broadcast operation. The filtering mechanism uses this预先 information to identify and transmit the broadcast only to the appropriate memory subset, thereby achieving effective partition isolation through advance preparation rather than complex real-time decision-making.

Inventive Principle:
Principle #10Preliminary action

4Adaptability or versatility

If dynamic reconfiguration of memory partition layouts is implemented, then adaptability is improved, but loss of time increases due to reconfiguration overhead

Engineering Contradiction:
Improvepartition layout flexibilityVSAvoidreconfiguration time
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent implements dynamics by enabling the memory partition layout to be dynamically reconfigured based on changing system requirements. The broadcast filtering mechanism adapts to new partition allocations and memory mappings in real-time, allowing the system to respond to varying workloads and partition needs without requiring system restarts or prolonged downtime.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent applies preliminary action by preparing and applying partition layout changes before they are needed by executing processes. The reconfiguration is performed in advance, allowing the system to be ready for new partition assignments without causing delays to active processes, thus minimizing the perceived reconfiguration time while maintaining high adaptability.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentUS20250272167A1Partition-aware broadcast operation filtering in a multiprocessor system
Publication Date: 2025.08.28 NVIDIA CORP
  • US20250272167A1 patent drawing
  • US20250272167A1 patent drawing
  • US20250272167A1 patent drawing

AI summary

Various embodiments include techniques for processing broadcast operations in a computing system. Typically, broadcast operations are transmitted to all portions of a particular subsystem, such as all cache slices in a cache memory. As a result, a process that issues a broadcast operation can interfere with one or more other processes that access portions of the subsystem not assigned to the process. To prevent such interference, logic in the computing system filters broadcast operations so as to transmit the broadcast operation to only the relevant portions assigned to the process that issued the broadcast operation. The logic tracks acknowledgments from the relevant portions and, when all pending acknowledgments have been received, the logic transmits a single acknowledgement to the process that issued the broadcast operation. The logic is dynamically configurable such that the logic can change the portions of the subsystem assigned to each process as needed.