Abstract

This publication describes a new user-behavior-guided dynamic loader in embedded operating systems (OS). It is well-understood that embedded devices, such as smartphones, are constrained in their random-access memory (RAM) resource. To better utilize the RAM resource, developers use shared libraries when building their application software (often referred to as apps or applications). Shared libraries reduce the memory footprint of individual application software. However, the utilization of shared libraries, even though beneficial in minimizing the volatile memory footprint, comes at a performance cost. The performance is improved, however, when shared libraries are pre-loaded before the application software is accessed by the end user. Current OS platforms lack the heuristic guided approach to predict which shared libraries are going to be needed at the time the end user accesses the application software. To this end, a new dynamic loader is developed to help predict the pre-loading of the needed shared libraries. To enable the OS to predict and pre-load shared libraries tailored to the end user, the new user-behavior-guided dynamic loader employs three components: user embedding, current time, and current location. To improve the performance of the dynamic loader, federated learning is utilized to democratize the computational power needed and benefit from each end user’s input data. By so doing, the described techniques optimize the prediction of the relevant shared libraries to be pre-loaded, while protecting the end user’s privacy. Consequently, user-behavior-guided dynamic loaders reduce the memory pressure of the embedded devices, while optimizing the performance of these devices.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.

Share

COinS