Aarushi Sarbhai

About Me

I am a second-year MS student at the School of Computing at University of Utah. I am advised by Prof. Sneha Kasera at the Advanced Networked Systems Research (ANSR) Lab. I am primarily interested in security and privacy problems. Currently, I am working on Privacy problems in Smart Homes.

I first got hooked on Computer Science in high school, at Delhi Public School, R.K. Puram, when I created the Marble Solitare game in C++. Then I went on to explore the area more through a bachelors degree at Manipal Institue of Technology. The areas that I was most excited to study were networking and data mining. During this time, through course projects and internships, I got real-world experience in web and app development.

When I am not sitting in front of a computing screen I usually take a break with a book or at the swimming pool. I love dancing and trained in Jazz dance for about 6 years. I also enjoy playing basketball and hiking.


Interests

Security Privacy Computer Networks Machine Learning


Affiliations

University of Utah

Research Assistant
M.S. Computer Science
August 2016 - May 2018

Manipal Institute of Technology

B.Tech. Computer Science and Engineering
August 2012 - July 2016

Oracle India Pvt. Ltd.

Project Intern
January 2016 - July 2016

University of Utah

Research Intern
June 2015 - July 2015

HelloTravel.com

App Development Intern
June 2014 - July 2014

Publication

  • Enabling WiFi in Open Access Networks
    HotWireless MobiCom 2017
  • The idea of open access networks is gradually becoming a reality with a large number of municipalities and communities deploying their own open network. In the open access network model, the municipality/community can act as the network operator and a multitude of services can be provided to the end users over the deployed infrastructure. In this age of wireless networks and mobile users, WiFi must also be an integral part of the open access network.

    The aspects of WiFi that need to be modified are identified along with the challenges that arise due to these modifications. These challenges are addressed by our simple, yet novel design for enabling WiFi in open access networks using Software Defined Networks (SDN) and access point virtualization. The access point can be dynamically configured to assign different services to different SSIDs using an Open-vSwitch (OVS) bridge in the access point. This OVS bridge is used to create mappings between the end users and the virtual ports. Using this approach, multiple users can be connected to a single access point and use their choice of service.

    This is the first attempt towards integrating open access networks and WiFi. A preliminary prototype of the design is implemented in the Emulab test bed to successfully verify its operation.

    Technologies Used: Emulab testbed, OpenFlow (RYU), Mininet

MS Thesis

  • Privacy-Aware Peak Load Reduction in Smart Homes
  • Power Grids across the world are getting “smarter” which need smart meters installed at each home to monitor power flow. These smart meters record power consumption data at every minute or even every second. This fine-grained data exposes private information about the residents of the house at many levels. Knowledge about the number of occupants, times of occupancy, appliance information and much more can be inferred from analyzing patterns in the electric usage.

    A solution to obscure this data is to add a battery to each home and use it strategically to manipulate the readings observed at the smart meter. The first set was to improve on the existing algorithms to control the battery. Deploying such a solution at a large scale can result in sudden peaks in the power demand. This is an alarming concern for the electric utility companies as they may not be able to prepare for these peaks. We propose a system to mitigate this problem while maintaining the privacy of the residents. This is achieved through a dynamic demand response mechanism that uses the battery to reduce peak load across a group of homes.

    The entire system is simulated in the open source power grid simulation tool, GridLAB-D to investigate the trade-off between the possible extent of data exposed and electric load reduced. Our algorithm to control the battery was able to improve privacy guarantees compared to other efforts in this direction while being able to reduce the peak consumption values.

    Technologues Used: GridLAB-D, Matlab

Course Projects

  • Automated DoS Mitigation with P4 enabled switches
  • Increased use of new SDN technologies like P4 that enable programmable data planes may cause potential attacks that are directed at this new exploit which was not possible before. In this work, I explored one such attack possible on P4-enabled switches directed at overloading the switches by targeting heavy processing tasks. I designed a basic framework to explore such attacks and investigate the effect of mitigation strategies. The framework is designed to be compatible with any P4 switch. It first monitors the rate of incoming packets. A high rate triggers the mitigation mechanism that identifies the heavy hitters based on their load. This information is used to drop the packets from these sources.

    Technologies Used: Python, Mininet, bmv2 software switches

  • Sending Dynamic Data Via Persistent Bluetooth Broadcast
  • The goal of this project was to study the loss in transmission of BLE broadcast packets in different spacial conditions. The implementation of the design involved building an Android app to receive packets and measure loss. Packets were transmitted by a Linux system using a Python wrapper. To decrease the packet loss in transmission, Forward Error Correction techniques were used with added redundancy. This algorithm resulted in minimal loss transmissions in ideal conditions.

    Technologies Used: Android, Python

  • Music Genre Classification and Feature Relation Identification
  • The aim of this project was to explore non-traditional features of songs, such as timbre, key, notes, etc, to classify music by genre. Insipred by the Kaggle challenge, We used the Million Song Dataset. We optimized a learning function to accurately classify music with approximately 60% accuracy using k-Nearest Neighbours and Support Vector Machine. This also helped in identifying the relationships between different kinds of features and genres.

    Technologies Used: Python

Applications in Use

  • Primavera Enterprise Project Portfolio Management
  • The Primavera EPPM Solution is a software package designed to enhance strategy execution, operational excellence, and financial performance across the enterprise. It is being used in numerous domains from IT corporations to Transport and Utility agencies across the world.

    During my internship at Oracle, I was tasked with designing the framework for implementing the security policies for the various products in the software suite. This enabled the business policies to be defined in software which could be easily modified according to the configurations of the clients using the software. Along with this module, I helped port the existing on-premise Primavera 17 software suite to the Oracle Public Cloud using the Chef infrastructure automation tool.

    Technologies Used: Chef

  • Student Placement Portal
  • I was a part of the Student Placement Portal team that built a platform to help automate the industrial recruitment process at Manipal Institute of Technology, Manipal. Users could upload edit information, create résumés, view results, etc. Companies can give information on the positions offered with detailed profiles, add student requirements, view student applications and other student statistics which is added by the administrative staff of the University Placement Office. Students, who met certain criteria, could register for the recruitment process. The platform is being actively used by thousands of students interested in the Internship and Full-Time Employment opportunities offered by different companies since 2015.

    Technologies Used: PHP, JS, CSS