Abstract:
The study used Message Passing Interface to build a novel approach for
forecasting memory leaks in HPC systems. Addressing resilient
challenges the introduction focused on HPC and Messages Processing
Infrastructure in system communication management. Throughout the
article, it was stressed that memory leaks must be found and fixed
rapidly to maintain system reliability. A greater inquiry was performed
on using machine learning to identify memory leaks in HPC systems.
The proposed approach involves collecting and preparing data from
MPI-based high-performance computing (HPC) systems, training trained
classification models for finding memory leak designs, and evaluating
model performance employing appropriate indicators. To predict, the
decision tree and Random forest algorithms, support vector machine
models, and the AdaBoost underwent analysis. The technique included
MPI metrics, feature engineering, and model training. Several
algorithms helped diagnose memory deficits, but Random Forest worked
best. The data and analysis showed these methods were helpful. The
study found that decision tree (DT) as well as random forest (RF)
algorithms could effectively diagnose and categories memory deficits
with near-perfect accuracy. Random Forest methods are consistently
better than baseline methods, with F-scores of 0.97 to 1.0 on the two
systems. Although the baselines were finished, their F-scores may range
from 0.89 to 0.97. The Random Forest technique comprised several
parameters, each of which affected classification accuracy. The Random
Forest approach selected the most important attributes using CPU and
memory use statistics.
