Showing papers by "Santonu Sarkar published in 2021"

Performance modeling of graphics processing unit application using static and dynamic analysis

Abstract: Graphics processing units (GPUs) have become an integral part of high‐performance computing to achieve an exascale performance. Understanding and estimating GPU performance is crucial for developers to design performance‐driven as well as energy‐efficient applications for a given architecture. This work presents a model developed using a static analysis of CUDA code to predict the execution time of NVIDIA GPU kernels without the need for running it. Here a PTX code is statically analyzed to extract instruction features, control flow, and data dependence. We propose a scheduling algorithm that satisfies resource reservation constraints to schedule these instructions in threads across streaming multiprocessors (SMs). We use dynamic analysis to build a set of memory access penalty models and use these models in conjunction with the scheduling information to estimate the execution time of the code. We present the experimental results which support that this approach works across architectures of NVIDIA GPUs. We first tested our model on two Kepler machines, where the mean percentage error (MPE)/mean absolute percentage error (MAPE) was − 8.88%/28.3% for Tesla K20 and − 5.66%/29.4% for Quadro K4200. We further tested the model on Maxwell and Pascal architectures and recorded the MPEs/MAPEs to be − 10.64%/47.8% and −3.94 %/28.5%, respectively.

d-BTAI: The Dynamic-Binary Tree Based Anomaly Identification Algorithm for Industrial Systems

Abstract: Many of the existing approaches to anomaly detection are based upon supervised learning and heavily dependent on training datasets. However, anomalies rarely occur in most industrial systems. Hence it is challenging to retrieve a training dataset labeled with true anomalies. Therefore, this motivates us to investigate such scenarios where it is arduous to get labeled data for anomalies. This paper has proposed a clustering-based recursive anomaly detection algorithm; dynamic-Binary Tree Anomaly Identifier (d-BTAI). d-BTAI has been applied on industrial devices since anomalies in large industrial devices can incur massive losses. The algorithm has experimented on various publicly available industrial datasets such as Cloudwatch, Yahoo, and Backblaze. d-BTAI has attained a higher Area under the ROC curve (AUC) in comparison with Isolation Forest (iForest), One Class Support Vector Machine (OCSVM), and Elliptic Envelope. The higher Negative Predictive Value (NPV) and specificity value demonstrate the algorithm’s efficacy on multiple datasets.

Understanding the Inter-Domain Presence of Research Topics in the Computing Discipline

Abstract: The very nature of scientific inquiry encourages the flow of ideas across research domains in a discipline. Research topics with higher inter-domain presence tend to attract higher attention at individual and organizational levels. This is more pronounced in a discipline like computing, with its deeply intertwined ideas and strong connections with technology. In this paper, we study corpora of research publications across four domains of the computing discipline – covering more than 150,000 papers, involving more than 200,000 authors over 55 years and 175 publication venues – to examine the influences on inter-domain presence of research topics. We find statistically significant evidence that higher collective eminence of researchers publishing on a topic is related to lower inter-domain presence of that topic, fewer authors publishing on a topic relate to the topic being likely to have higher inter-domain presence, while topics belonging to more close-knit clusters of topics are likely to have lower inter-domain presence. Our results can inform decisions around defining and sustaining research agendas and offer insights on the progression of the computing discipline.