Multi-instance LSTM RNN Memory Page Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modern datacenter memory pools face challenges in accommodating diverse demands due to increasingly distributed and heterogeneous memory hierarchies, which are difficult to manage efficiently across various applications and workloads.
Innovation Solution
A hybrid memory management system using machine learning-based techniques, specifically implementing multiple instances of long short-term memory (LSTM) recurrent neural networks (RNNs) to predict memory page access patterns and migrate pages to faster memory types, while also utilizing a default prediction scheme for computationally expensive operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional memory management schemes are used, then system simplicity is maintained, but flexibility and adaptability to diverse applications and memory access patterns deteriorate
Solution Approach 1:
The patent segments the memory management system by introducing multiple specialized memory predictors (e.g., LRU predictor, LSTM-based predictor) that operate in parallel. Each predictor handles specific types of memory access patterns, allowing the system to adapt to diverse applications without requiring a complete redesign of the memory management architecture. This segmentation enables targeted optimization for different workload characteristics while maintaining overall system manageability.
2Measurement precision
If machine learning-based prediction is applied to all memory pages, then prediction accuracy improves, but computational overhead and energy consumption increase
Solution Approach 1:
The patent implements partial action by applying machine learning-based prediction selectively rather than universally. It introduces a hybrid approach where simple predictors (like LRU) handle the majority of memory pages, while ML-based predictors are applied only to specific pages or workloads that benefit most from their predictive capabilities. This selective application reduces computational overhead and energy consumption while maintaining high prediction accuracy for critical memory access patterns.
3Speed
If memory pages are migrated to faster memory types, then access speed improves, but energy consumption for data transfer increases
Solution Approach 1:
The patent applies preliminary action by predicting future memory access patterns and proactively migrating memory pages to faster memory types before they are actually needed. The ML-based predictors analyze historical access patterns and anticipate which pages will be accessed soon, allowing the system to pre-position these pages in high-speed memory. This eliminates the need for costly emergency migrations when pages are actually accessed, reducing overall energy consumption while maintaining fast access speeds.
Data Source
AI summary
A computer processing system having a first memory with a first set of memory pages resident therein and a second memory coupled to the first memory. A resource tracker provides information to instances of a long short-term memory (LSTM) recurrent neural network (RNN). A predictor identifies memory pages from the first set of memory pages for prediction by the one or more LSTM RNN instances. The system groups the memory pages of the identified plurality of memory pages into a number of patterns based on a number of memory accesses per time. An LSTM RNN instance predicts a number of page accesses for each pattern. A second set of memory pages is selected for moving from the first memory to the second memory.