ML-Based Data Loading for Cloud Database Latency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Database systems face performance issues due to high latency when accessing frequently accessed 'hot' data, leading to inefficient data swapping between memory types during workload execution, which negatively impacts system performance and resource utilization.
Innovation Solution
Implement a data loading system that uses machine learning models to predict workload patterns and selectively load 'hot' data into low-latency memory based on predicted access frequencies, thereby reducing latency and conserving resources by minimizing data swapping.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If data is stored in low-latency memory to reduce access latency, then access speed improves, but memory capacity is limited and data swapping frequency increases
Solution Approach 1:
The system performs preliminary actions by predicting which data will be accessed soon and pre-loading it into low-latency memory before actual access occurs. This proactive approach ensures hot data is already in fast memory when needed, reducing access latency without causing excessive swapping, as the loading is done in advance based on predictions rather than reactive swapping.
Solution Approach 2:
The system dynamically adjusts the data loading strategy by using machine learning models to predict workload patterns and adaptively determine which data to load into memory. The prediction accuracy and loading decisions are continuously refined based on historical workload data, allowing the system to optimize the balance between memory utilization and access speed according to changing access patterns.
2Loss of time
If all data is loaded into low-latency memory, then access latency is minimized, but resource utilization decreases due to excessive data swapping
Solution Approach 1:
Instead of loading all data into low-latency memory, the system applies partial action by selectively loading only the predicted hot data that will be accessed soon. This approach minimizes memory swapping operations and the associated energy consumption while still achieving low latency for the critical subset of frequently accessed data, rather than wasting resources loading cold data that won't be accessed.
Solution Approach 2:
The system changes the parameter of data selection from static (all data or random data) to dynamic (predicted hot data based on workload patterns). By using machine learning models to analyze historical workload data and predict future access patterns, the system transforms the data loading decision into an adaptive process that optimizes both latency reduction and energy efficiency by loading only the necessary portion of data.
3Measurement precision
If machine learning models predict workload patterns accurately, then data loading precision improves, but system complexity increases
Solution Approach 1:
The machine learning model serves multiple functions: it predicts workload patterns, identifies hot data, and guides data loading decisions. This multi-functionality reduces the need for separate specialized components for each task, thereby managing system complexity while achieving accurate workload prediction and effective data loading strategies through a single integrated predictive system.
Data Source
AI summary
Methods, systems, and computer-readable storage media for receiving a workload period, during which a workload is applied to a database system, providing a set of ML models based on historical data representative of historical executions of the workload over the workload period, each ML model configured to predict a cluster arrival rate curve (cARC), and during execution of the workload period and, for each timeslice of a plurality of timeslice of the workload period: providing a predicted cARC from each ML model, the predicted cARC representative of a predicted workload, determining column visiting times for each of a plurality of columns of each of a plurality of tables stored in the database system, generating a column list based on the column visiting times, and loading column data representative of columns included in the column list into low-latency memory prior to execution of a workload during the respective timeslice.