Proactive Working Memory Management via Predictive Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing memory management approaches for computing nodes often lead to wasteful and costly operations, including memory exhaustion, which results in 'page-out' of transaction data from working memory into storage memory, requiring costly recovery processes.
Innovation Solution
A proactive memory management system that predicts future state data using predictive models to proactively establish control parameter values for working memory, allowing for efficient execution of multiple concurrent transactions by reserving memory space and pausing or holding transactions as necessary to prevent memory exhaustion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional memory management approaches are used, then simple memory allocation is maintained, but memory exhaustion occurs leading to page-out and costly recovery processes
Solution Approach 1:
The system performs preliminary actions by predicting future transaction requirements and proactively allocating memory resources before transactions actually occur. The predictive model analyzes historical data to forecast which transactions will be invoked and how much memory they will require, allowing the memory manager to reserve appropriate memory space in advance, thereby preventing memory exhaustion and avoiding costly page-out recovery operations.
Solution Approach 2:
The system implements feedback mechanisms by continuously monitoring actual transaction behavior and comparing it against predictions. The memory manager uses feedback from actual memory usage patterns to refine and update the predictive models, improving accuracy over time. This feedback loop allows the system to learn from real-world usage and enhance its memory management effectiveness progressively.
2Productivity
If proactive memory management with predictive models is implemented, then memory exhaustion is avoided and transaction execution efficiency is enhanced, but system complexity increases
Solution Approach 1:
The memory management system performs self-service by autonomously making predictions about future transaction needs and automatically executing memory allocation decisions without requiring manual intervention. The predictive models self-update based on observed patterns, and the memory manager self-regulates resource distribution, reducing the need for external control and simplifying operational complexity despite the advanced functionality.
Solution Approach 2:
By performing memory allocation decisions in advance based on predictions, the system avoids the complexity of reactive memory management. Transactions are allocated memory resources before they are invoked, eliminating the need for complex real-time memory arbitration and recovery operations that would otherwise be required when memory exhaustion occurs.
3Adaptability or versatility
If memory space is reserved proactively for predicted transactions, then concurrent transaction execution is improved, but memory allocation decisions become more complex
Solution Approach 1:
The system resolves allocation complexity by making decisions in advance. The predictive model evaluates multiple potential transactions and their memory requirements beforehand, allowing the memory manager to pre-determine allocation decisions. This eliminates the need for complex real-time allocation algorithms and enables smooth handling of concurrent transactions through predetermined memory reservations.
Solution Approach 2:
The system manages complexity by dynamically adjusting prediction parameters and allocation strategies based on system state. The predictive models can change their prediction horizons, accuracy thresholds, and allocation aggressiveness parameters adaptively, allowing the system to handle varying concurrent transaction workloads with a single unified approach rather than requiring complex specialized logic for different scenarios.
Data Source
AI summary
Methods, computer program products, and systems are presented. The method computer program products, and systems can include, for instance: predicting, with use of one or more predictive model, subsequent future state data of a computing node having an associated working memory, wherein the subsequent future state data specifies that a certain transaction currently uninvoked will be invoked by the computing node; prior to the invoking of the certain transaction, proactively establishing one or more control parameter value for controlling the working memory in dependence on the future state data; invoking the certain transaction in response to receipt, by the computing node, of transaction invoking request data for invoking the certain transaction; and executing the certain transaction in dependence on at least one control parameter value of the one or more control parameter value for controlling the working memory.