Prediction and optimization of memory fragmentation based on eBPF and LightGBM in high-load environments

With the rapid development and widespread application of embedded systems, ensuring the security and reliability of data storage has become increasingly important. Under high load environments, resource-constrained embedded systems frequently allocate and deallocate memory, leading to significant memory fragmentation that impacts system performance. To address this issue, this paper proposes a memory fragmentation prediction model based on eBPF and LightGBM, and applies it to embedded systems under high load environment. The model utilizes eBPF programs to collect key performance indicators of the system memory, including fine-grained data such as physical page usage. By constructing datasets and performing preprocessing and cleaning operations, this paper trains the LightGBM model with various parameter combinations to build an accurate memory fragmentation prediction model. This prediction model can accurately predict future memory fragmentation levels in the system and implement regular defragmentation measures when necessary to prevent excessive fragmentation, thereby improving the success rate of memory allocations. Experimental results demonstrate that, under high load conditions, the memory fragmentation prediction model based on eBPF and LightGBM performs excellently in predicting memory fragmentation in embedded systems. The model not only accurately predicts future fragmentation levels but also contributes to memory management and optimization in embedded systems, enhancing system stability and performance.