Hybrid Local Cache Model for Enhancing Performance and Memory Footprint of Mobile Heap Allocator

The representative mobile heap allocator that has been used in Android system, called SCUDO, supports two local cache models: the shared model and the exclusive model. However, neither the shared model nor the exclusive model cannot satisfy both memory usage and performance requirements simultaneously since mobile applications have diverse memory allocation/free patterns and the number of threads varies from single to several tens. This paper suggests a hybrid local cache model that aims to adaptively meet memory usage and performance goals. Our evaluation results show that our hybrid local cache model outperforms the shared model by up to 22% on allocation time, while preserving the memory usage.