Workload DAG Mapping Across Memory Tiers and Connectivity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional workload performance systems fail to account for memory tier characteristics and processing system/memory system connectivity, which affect the ability to satisfy workload requirements.
Innovation Solution
A resource management system that determines resource capabilities and connectivity between processing and memory systems, identifies a Directed Acyclic Graph (DAG) for workload functions, and configures optimal processing system/memory system/connectivity combinations to perform those functions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional workload performance systems use distributed processing systems and memory fabrics with multiple memory tiers, then the system can provide varied memory capabilities (persistence, latency, durability, capacity), but the system cannot deterministically satisfy workload requirements due to non-deterministic connectivity influence on memory tier behavior
Solution Approach 1:
The patent changes the parameters considered for resource allocation by incorporating connectivity metrics (bandwidth, latency, reliability) alongside traditional memory tier characteristics. This allows the system to predictively determine memory tier behavior under specific connectivity conditions, transforming non-deterministic behavior into predictable outcomes that can satisfy workload requirements.
Solution Approach 2:
The system implements feedback mechanisms that monitor actual connectivity performance and use this information to adjust workload placement decisions. By continuously gathering data on memory tier behavior under different connectivity conditions and using this feedback to refine allocation strategies, the system achieves more reliable workload satisfaction while maintaining adaptability.
2Productivity
If the system considers multiple memory tiers with different characteristics, then the system can optimize for different workload requirements, but the complexity of determining optimal resource allocation increases due to connectivity variables
Solution Approach 1:
The patent segments the resource allocation problem into manageable components by evaluating connectivity characteristics (bandwidth, latency, reliability) as separate, independent factors that can be assessed and weighted individually. This segmentation allows the system to process complex multi-tier memory decisions through a structured approach that considers each connectivity dimension separately, reducing overall computational complexity.
Solution Approach 2:
The system transforms the complex allocation problem by changing the parameter set to include connectivity metrics alongside memory tier characteristics. This parameter expansion creates a more comprehensive but systematically organized evaluation framework that handles complexity through structured parameter assessment rather than unstructured analysis.
3Adaptability or versatility
If the system uses non-deterministic connectivity between processing systems and memory systems, then the system architecture remains flexible, but the ability to predictively satisfy workload requirements deteriorates
Solution Approach 1:
The patent employs feedback loops that continuously measure actual connectivity performance and use this information to refine predictions of memory tier behavior. By gathering real-world data on how connectivity conditions affect memory access patterns and using this feedback to improve prediction models, the system maintains architectural flexibility while enhancing prediction accuracy.
Solution Approach 2:
The system performs preliminary assessment of connectivity characteristics before making workload allocation decisions. By evaluating bandwidth, latency, and reliability metrics in advance and using this preliminary information to predict memory tier behavior, the system can make more accurate workload requirement predictions while preserving architectural flexibility for future adjustments.
Data Source
AI summary
The present disclosure describes a system that includes a resource management system that is coupled to a plurality of processing systems and a plurality of memory systems. The resource management system determines resource capabilities provided by each of the plurality of processing systems, each of the plurality of memory systems, and connectivity between the plurality of processing systems and the plurality of memory systems. When the resource management system receives a workload request to perform a first workload, it identifies a first Directed Acyclic Graph (DAG) that includes a plurality of functions for performing the first workload and, for each of the plurality of functions, determines a processing system/memory system/connectivity combination based on the resource capabilities that provides a function performance capability that satisfies a function requirement for that function, and configures that respective processing system/memory system/connectivity combination to perform that function.