Media Playback Memory Prediction to Prevent Out-of-Memory Kills
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Media streaming applications often experience unpredictable 'out of memory kills' due to unknown device-specific memory thresholds, leading to application crashes and poor user experience, as manufacturers do not publicly disclose these limits.
Innovation Solution
A machine learning model is trained using static and dynamic device information to predict out of memory kills, implementing a graded window to indicate likelihood within a specified timeframe, and adjusts memory usage to prevent crashes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If media streaming applications run on playback devices with limited memory, then the application can operate on a wider range of devices, but the application is subject to out of memory kills and crashes
Solution Approach 1:
The system performs preliminary actions by collecting static device information and dynamic playback information before crashes occur, trains a machine learning model to predict memory threshold characteristics, and proactively adjusts streaming buffer sizes to prevent out of memory kills, thereby maintaining application stability across devices with limited memory
Solution Approach 2:
The system changes parameters by dynamically adjusting the streaming buffer size based on predicted device memory thresholds. The machine learning model determines optimal buffer sizes by analyzing device characteristics and playback conditions, allowing the application to adapt its memory usage parameters to avoid crashes while maintaining compatibility across diverse devices
2Object-affected harmful factors
If device-specific memory thresholds are not disclosed by manufacturers, then device security and control are maintained, but media providers cannot predict when applications will crash
Solution Approach 1:
The system implements feedback by collecting playback information including crash data from multiple devices, using this feedback to train and refine the machine learning model, which then predicts memory threshold characteristics for similar devices. This closed-loop feedback mechanism enables the system to learn and predict crash conditions without requiring manufacturers to disclose proprietary memory threshold information
Solution Approach 2:
The machine learning model serves as an intermediary that infers memory threshold characteristics indirectly through pattern recognition in aggregated playback data. Instead of directly accessing or relying on undisclosed manufacturer specifications, the model mediates between observable playback behaviors and predicted crash conditions, enabling predictions without compromising device security or obtaining proprietary information
3Duration of action of stationary object
If the streaming buffer size is increased to improve playback continuity, then buffer underruns are reduced, but the application uses more memory and risks out of memory kills
Solution Approach 1:
The system applies dynamics by making the streaming buffer size adaptive rather than static. The machine learning model continuously evaluates device-specific memory thresholds and playback conditions to dynamically adjust the buffer size, allowing the application to optimize playback continuity on devices with sufficient memory while automatically reducing buffer sizes on memory-constrained devices to prevent crashes
Data Source
AI summary
A computer-implemented method includes identifying static information associated with a computing device that is running a media playback application. The method also includes monitoring the media playback application during a playback session to identify dynamic information associated with the playback session. Still further, the method includes instantiating a trained machine learning (ML) model to determine, based on historical usage data associated with the media playback application, a likelihood that the media playback application will experience an out of memory kill. The trained ML model implements a graded window to indicate a scaled likelihood that an out of memory kill will occur within a specified timeframe. Then, according to the trained ML model's determination, the method generates a prediction that an out of memory kill will occur for the media playback application within the specified timeframe. Various other methods, systems, and computer-readable media are also disclosed.