Fabric-Attached Memory Allocation by Dataset Security Threshold
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High Performance Computing (HPC) applications generate large numbers of nearly congruent datasets over extended run-times, where minute differences across datasets may be analytically significant, necessitating secure and efficient management of dynamic memory resources to handle datasets with varying security thresholds and lifetimes.
Innovation Solution
Implementing a memory manager that allocates secure and non-secure memory regions, utilizing encryption and data obfuscation methods based on data-oriented security ranking values, and managing memory lanes with configurable encryption thresholds to store datasets with varying levels of security and persistence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If datasets are encrypted and stored in secure memory regions, then data security is improved, but memory access time and processing speed deteriorate
Solution Approach 1:
The memory system is segmented into multiple memory lanes (secure and non-secure regions) with different security characteristics. Datasets are allocated to specific lanes based on their security requirements, allowing simultaneous encrypted and unencrypted access paths. This segmentation enables parallel processing where security-critical data goes through encryption/decryption paths while non-critical data accesses directly, resolving the contradiction between security and speed.
Solution Approach 2:
Different security levels and access policies are applied to different regions of memory based on local requirements. Each memory lane can have its own encryption configuration, access controls, and performance characteristics. This local quality approach allows the system to optimize for both security and speed in different locations simultaneously, rather than applying a uniform security policy across all memory.
2Reliability
If memory is allocated for extended periods to preserve dataset integrity, then data reliability is improved, but memory resource utilization deteriorates
Solution Approach 1:
The memory allocation system dynamically adjusts dataset lifetimes and memory retention periods based on analytical significance, security requirements, and current system state. Rather than static long-term allocation, the system can extend or terminate memory residency of datasets as needed, allowing previously allocated memory to be reused while maintaining integrity for datasets that still require it. This dynamic approach resolves the contradiction between reliability and resource utilization.
Solution Approach 2:
The system changes key parameters such as memory retention time, security level, and priority based on dataset characteristics and system conditions. Datasets with high analytical significance maintain longer memory residence, while less critical datasets are moved to storage or deallocated. This parameter-based management enables flexible resource allocation that maintains reliability where needed while improving overall productivity.
3Measurement precision
If multiple security thresholds are implemented for different datasets, then security precision is improved, but system complexity deteriorates
Solution Approach 1:
The memory management system automatically determines appropriate security thresholds and memory lane allocations based on dataset metadata, security policies, and system configuration, without requiring manual intervention for each dataset. The system self-manages the complexity of multiple security levels by implementing automated classification, allocation, and access control mechanisms that adapt to different security requirements while maintaining manageable system operation.
Data Source
AI summary
Systems and methods are provided for secure data subsets in a memory-centric computer system. A method includes receiving, in a computer system, a request for allocation of a region of a memory. The request includes a data-oriented security ranking value associated with a dataset to be stored in the region of memory. The method further includes comparing the data-oriented security ranking value to a first security threshold. In response to determining that the data-oriented security ranking value meets or exceeds the first security threshold, the method includes encrypting the dataset using an encryption key and allocating a region of memory in a portion of the memory reserved for encrypted data.