My Research

I am interested in Cloud Computing, Serverless computing and Machine Learning. I am working on optimal resource utilization, resource allocation and workload for Cloud Computing and performance optimization of machine learning systems.

Machine Learning Training in a Serverless Environment – Fall 2019 to spring 2021

• Adviser: Dr. Feng Yan
• Collaboration: Bell Labs
• Student lead of project, working on scalable Distributed Machine Learning training in a serverelss environemnt. The goal of the project is to minimize the overall training time, monerty cost, resource provisioninig overhead and communication overhead during the gradient updates.

Machine Learning Inference in a Serverless Environment – Summer 2019 to Summer 2020

• Adviser: Dr. Feng Yan
• Collaboration: College of William & Marry
• Student lead of project, working on an optimal Configuration allocation (Memory allocation, batch size and batch timeout) to meet the SLO while minimizing the cost for real time inference in a serverless environment.

Efficient Client Selection Policy in a Federaled Learning Environment – Spring 2019 to Fall 2019

• Adviser: Dr. Feng Yan
• Collaboration: IBM Research and George Mason University
• Proposed a client selection policy to handle the stragglers probelm taking into consideration the response time of a client, total training time and accuracy of a model in a Federated Learning Environment

The Most Profitable use of Burstable Instances of Amazon EC2 – Spring 2017 to Summer 2019

• Adviser: Dr. Feng Yan
• Collaboration: College of William & Marry
• Proposed an efficient technique using light weight profiling and quantile regression which yields a win-win for both cloud service provider and consumer (Student lead in the project).

Cooling Cost Optimization of Geographically Distributed Data Centres Spring 2016 to Fall 2016

• Adviser: Dr. Oznur Ozkasap
• The cooling cost and network delays of geographically distributed data centers are minimized by using inter/intra data center scheduling techniques. To achieve cooling cost optimization and delay minimization, three different workload scheduling techniques are proposed. These techniques are implemented in Java using CloudSim simulator by adding time varying characteristics to data centers. (CloudSim)

Cooling Cost Optimization of Geographically Distributed Data Centres Fall 2015

• Adviser: Dr. Oznur Ozkasap
• The state of the art scheduling and electricity cost optimization techniques for geographically distributed data centers is analyzed. The techniques are divided into 4 major categories based on their similarities and cost model is proposed.

Environment-Friendly Energy Efficient Distributed Data Centers Fall 2014- Spring 2015

• Adviser: Dr. Oznur Ozkasap
• Data center cooling cost and the SLA violations are minimzed by scheduling the real world workload based on the temperature variation within the data center and the user specified deadlines of the requests.