Shoney Sebastian
Research Subjects
Shoney Sebastian has a total of 11 co-authors. Here are the top co-authors by number of shared publications:
- Josephine Eskaline Joyce
IBM India Private Limited,Bangalore,India
2 shared publications - Diniya Jose
Deemed to be University
1 shared publication - Hikmat A. M. Abdeljaber
Department of Computer Science, Faculty of Information Technology, Applied Science Private University, Amman, Jordan
1 shared publication - Iyyappan Moorthi
Ahlia University, Bahrain
1 shared publication - Jabeen Nazeer
Department of Computer Science, College of Computer Engineering and Sciences, Prince Sattam Bin Abdulaziz University, Alkharj, Saudi Arabia
1 shared publication - Joseph K. Iype
CHRIST (Deemed to be University)
1 shared publication
Latest Publications
- Open Access
- 10 June 2026
- Published by MDPI AG in Technologies
- Vol. 14 (6) , 350
Inference-Time-Driven Autoscaling for Inference Workloads: A Comparative Study of Latency-Variant Models in Kubernetes
Kubernetes Horizontal Pod Autoscaler (HPA) primarily relies on resource-based metrics, such as CPU utilization, which are poorly suited to capturing the latency variability of AI inference workloads. In this paper, we propose a custom-metric-driven autoscaling approach that leverages inference latency histograms as first-class scaling signals for Kubernetes HPA. The proposed framework integrates a Prometheus Operator (PO)-based observability stack with the Prometheus Adapter to expose and aggregate per-pod inference latency metrics, enabling workload-aware scaling decisions. We evaluate the approach using four mid-scale transformer-based inference services, comprising two reasoning-like and two latency-stable workloads, under high-concurrency conditions. The experiments analyze latency variation, tail behavior, and replica dynamics across multiple autoscaling policies, including variations in scale-up aggressiveness (3 pods/30 s, 3 pods/60 s, 6 pods/60 s), inference-time thresholds, and stabilization windows. Compared to CPU-based autoscaling, inference-driven policies reduce mean response time by 18–27% for reasoning-like workloads and 12–20% for stable workloads. The results show that latency-variable workloads exhibit wider tails and higher variance, indicating the need for moderately aggressive scale-up strategies to avoid long-lasting latency spikes. Overall, the findings show that inference-latency-driven custom metrics significantly improve autoscaling efficiency and stability for transformer-based inference workloads in cloud-native environments.
- 10 June 2026
- Published by MDPI AG in Technologies
- Vol. 14 (6) , 350
Open AccessInference-Time-Driven Autoscaling for Inference Workloads: A Comparative Study of Latency-Variant Models in Kubernetes
Kubernetes Horizontal Pod Autoscaler (HPA) primarily relies on resource-based metrics, such as CPU utilization, which are poorly suited to capturing the latency variability of AI inference workloads. In this paper, we propose a custom-metric-driven autoscaling approach that leverages inference latency histograms as first-class scaling signals for Kubernetes HPA. The proposed framework integrates a Prometheus Operator (PO)-based observability stack with the Prometheus Adapter to expose and aggregate per-pod inference latency metrics, enabling workload-aware scaling decisions. We evaluate the approach using four mid-scale transformer-based inference services, comprising two reasoning-like and two latency-stable workloads, under high-concurrency conditions. The experiments analyze latency variation, tail behavior, and replica dynamics across multiple autoscaling policies, including variations in scale-up aggressiveness (3 pods/30 s, 3 pods/60 s, 6 pods/60 s), inference-time thresholds, and stabilization windows. Compared to CPU-based autoscaling, inference-driven policies reduce mean response time by 18–27% for reasoning-like workloads and 12–20% for stable workloads. The results show that latency-variable workloads exhibit wider tails and higher variance, indicating the need for moderately aggressive scale-up strategies to avoid long-lasting latency spikes. Overall, the findings show that inference-latency-driven custom metrics significantly improve autoscaling efficiency and stability for transformer-based inference workloads in cloud-native environments.
- 6 February 2025
- Published by Springer Nature
- p. 31-44
Maximizing Efficiency: Unveiling the Potential of Kubernetes Metrics
- 6 February 2025
- Published by Springer Nature
- p. 31-44
Maximizing Efficiency: Unveiling the Potential of Kubernetes Metrics
- 31 August 2024
- Published by Taylor & Francis in International Journal of Computers and Applications
- Vol. 46 (10) , 921-932
Elastic circuit de-constructor: a pattern to enhance resiliency in microservices
- 31 August 2024
- Published by Taylor & Francis in International Journal of Computers and Applications
- Vol. 46 (10) , 921-932
Elastic circuit de-constructor: a pattern to enhance resiliency in microservices
- Open Access
- 26 April 2024
- Published by Wiley in Wireless Communications and Mobile Computing
- Vol. 2024, 1-9
Hybrid Agent-Based Load-Balancing Approach Used in an IaaS Platform
Cloud computing has received a resounding welcome. It was created following methodical and thorough study in web services, distributed computing, utility computing, and virtualization, and it offers several benefits, including lower costs, less space required, and easier management. These advantages bring in a significant number of new users to the cloud platform every day. In addition, because cloud computing is an Internet-based computing paradigm, it must deal with the issue of overwhelming demands through effective load-balancing. A very small number of studies only focus on load-balancing problems in cloud computing platforms, while the majority of load-balancing research is accessible in many domains, including parallel, distributed, and grid computing. Infrastructure as a Service (IaaS), Software as a Service, and Platform as a Service are the three basic categories under which cloud computing falls. For these models, there are notable differences in the load-balancing techniques used. This work compared the outcome with the current method and presented a hybrid agent-based load-balancing approach for the IaaS platform.
- 26 April 2024
- Published by Wiley in Wireless Communications and Mobile Computing
- Vol. 2024, 1-9
Open AccessHybrid Agent-Based Load-Balancing Approach Used in an IaaS Platform
Cloud computing has received a resounding welcome. It was created following methodical and thorough study in web services, distributed computing, utility computing, and virtualization, and it offers several benefits, including lower costs, less space required, and easier management. These advantages bring in a significant number of new users to the cloud platform every day. In addition, because cloud computing is an Internet-based computing paradigm, it must deal with the issue of overwhelming demands through effective load-balancing. A very small number of studies only focus on load-balancing problems in cloud computing platforms, while the majority of load-balancing research is accessible in many domains, including parallel, distributed, and grid computing. Infrastructure as a Service (IaaS), Software as a Service, and Platform as a Service are the three basic categories under which cloud computing falls. For these models, there are notable differences in the load-balancing techniques used. This work compared the outcome with the current method and presented a hybrid agent-based load-balancing approach for the IaaS platform.
- 3 February 2024
- Published by Springer Nature
- p. 222-234
Comparative Analysis of State-of-the-Art Face Recognition Models: FaceNet, ArcFace, and OpenFace Using Image Classification Metrics
- 3 February 2024
- Published by Springer Nature
- p. 222-234
Comparative Analysis of State-of-the-Art Face Recognition Models: FaceNet, ArcFace, and OpenFace Using Image Classification Metrics
- Open Access
- 1 October 2019
- Published by Institute of Advanced Engineering and Science in International Journal of Electrical and Computer Engineering (IJECE)
- Vol. 9 (5) , 3927-3933
Taylor series method in TDOA approach for indoor positioning system
Localisation technologies have always remained in the limelight of positioning-science as researchers have ever shown keen interest to know the exact positions of things. Ultrasonic sensors are mainly used for localisation of mobile robots since they provide high accuracy. This paper presents Taylor Series Method in Time Difference of Arrival approach using ultrasonic sensors.Signals are send from the sensors periodically.The time difference of arrival of signals from the ultrasonic sensors is used by the receiver unit to estimate the location of the mobile unit. The equations formed by using Time Difference of Approach are solved using Taylor Series Method. Taylor Series Method provides a more accurate result since they give less error compared to other methods and they ignore the measurement errors.
- 1 October 2019
- Published by Institute of Advanced Engineering and Science in International Journal of Electrical and Computer Engineering (IJECE)
- Vol. 9 (5) , 3927-3933
Open AccessTaylor series method in TDOA approach for indoor positioning system
Localisation technologies have always remained in the limelight of positioning-science as researchers have ever shown keen interest to know the exact positions of things. Ultrasonic sensors are mainly used for localisation of mobile robots since they provide high accuracy. This paper presents Taylor Series Method in Time Difference of Arrival approach using ultrasonic sensors.Signals are send from the sensors periodically.The time difference of arrival of signals from the ultrasonic sensors is used by the receiver unit to estimate the location of the mobile unit. The equations formed by using Time Difference of Approach are solved using Taylor Series Method. Taylor Series Method provides a more accurate result since they give less error compared to other methods and they ignore the measurement errors.
- Open Access
- 1 April 2019
- Published by Institute of Advanced Engineering and Science in International Journal of Informatics and Communication Technology (IJ-ICT)
- Vol. 8 (1) , 56-62
Predicting heart ailment in patients with varying number of features using data mining techniques
Data mining can be defined as a process of extracting unknown, verifiable and possibly helpful data from information. Among the various ailments, heart ailment is one of the primary reason behind death of individuals around the globe, hence in order to curb this, a detailed analysis is done using Data Mining. Many a times we limit ourselves with minimal attributes that are required to predict a patient with heart disease. By doing so we are missing on a lot of important attributes that are main causes for heart diseases. Hence, this research aims at considering almost all the important features affecting heart disease and performs the analysis step by step with minimal to maximum set of attributes using Data Mining techniques to predict heart ailments. The various classification methods used are Naïve Bayes classifier, Random Forest and Random Tree which are applied on three datasets with different number of attributes but with a common class label. From the analysis performed, it shows that there is a gradual increase in prediction accuracies with the increase in the attributes irrespective of the classifiers used and Naïve Bayes and Random Forest algorithms comparatively outperforms with these sets of data.
- 1 April 2019
- Published by Institute of Advanced Engineering and Science in International Journal of Informatics and Communication Technology (IJ-ICT)
- Vol. 8 (1) , 56-62
Open AccessPredicting heart ailment in patients with varying number of features using data mining techniques
Data mining can be defined as a process of extracting unknown, verifiable and possibly helpful data from information. Among the various ailments, heart ailment is one of the primary reason behind death of individuals around the globe, hence in order to curb this, a detailed analysis is done using Data Mining. Many a times we limit ourselves with minimal attributes that are required to predict a patient with heart disease. By doing so we are missing on a lot of important attributes that are main causes for heart diseases. Hence, this research aims at considering almost all the important features affecting heart disease and performs the analysis step by step with minimal to maximum set of attributes using Data Mining techniques to predict heart ailments. The various classification methods used are Naïve Bayes classifier, Random Forest and Random Tree which are applied on three datasets with different number of attributes but with a common class label. From the analysis performed, it shows that there is a gradual increase in prediction accuracies with the increase in the attributes irrespective of the classifiers used and Naïve Bayes and Random Forest algorithms comparatively outperforms with these sets of data.
- 1 January 2017
- Published by Diva Enterprises Private Limited in International Journal of Applied Research on Information Technology and Computing
- Vol. 8 (3) , 324-334
A Load Balancing Algorithm for Infrastructure as a Service Platform (IaaS) Using Bees Algorithm and Queuing Theory
Cloud computing is an emerging computing paradigm in the IT industry and the acceptance of this technology is rising fast with an unprecedented growth of data centre across the world. Recent studies have shown that ensuring load balancing (LB) in different data centre is critical for achieving optimal resource utilisation, maximise throughput, minimum response time and avoid overloading of resource. The traditional techniques of LB are not suitable for the cloud data centre's because of the virtualisation of the physical systems. In a centralised approach, scalability issues are more especially on machines with small amount of memory. On the other hand, a fully distributed approach tends to yield a poor performance on very large number of systems. In this paper, we propose a hybrid agent-based LB algorithm for cloud computing using Bees algorithm and queuing theory for Infrastructure as a Service platform. The proposed algorithm solves the problem of load unbalance and ensures high resource utilisation with minimum response time.
- 1 January 2017
- Published by Diva Enterprises Private Limited in International Journal of Applied Research on Information Technology and Computing
- Vol. 8 (3) , 324-334
A Load Balancing Algorithm for Infrastructure as a Service Platform (IaaS) Using Bees Algorithm and Queuing Theory
Cloud computing is an emerging computing paradigm in the IT industry and the acceptance of this technology is rising fast with an unprecedented growth of data centre across the world. Recent studies have shown that ensuring load balancing (LB) in different data centre is critical for achieving optimal resource utilisation, maximise throughput, minimum response time and avoid overloading of resource. The traditional techniques of LB are not suitable for the cloud data centre's because of the virtualisation of the physical systems. In a centralised approach, scalability issues are more especially on machines with small amount of memory. On the other hand, a fully distributed approach tends to yield a poor performance on very large number of systems. In this paper, we propose a hybrid agent-based LB algorithm for cloud computing using Bees algorithm and queuing theory for Infrastructure as a Service platform. The proposed algorithm solves the problem of load unbalance and ensures high resource utilisation with minimum response time.