Rajrup Ghosh

Ph.D. Candidate, USC@NSL

  • (2019- ) Ph.D. in Computer Science, University of Southern California, USC.
  • (2017) M.Tech in Computational Science, Indian Institute of Science (IISc), Bangalore.
  • (2015) B.E. in Computer Science, Indian Institute of Engineering Science and Technology (IIEST), Shibpur.


I am a Ph.D. student in Networked System Lab (NSL) at University of Southern California. I am fortunate to be advised by Prof. Ramesh Govindan. My primary research interests are in areas of Volumetric Video, Adaptive Video Streaming, Edge+Cloud Computing and Systems for Machine Learning.

Prior to joining USC, I completed my Masters in Computational Science at the Department of Computational and Data Sciences (CDS), Indian Institute of Science (IISc), Bangalore. I was advised by Prof. Yogesh Simmhan at DREAM:Lab.

Work Experience

Graduate Research Assistant (2019 - present)
Department of Computer Science, University of Southern California.

Research Intern (June 2020 - August 2020)
Microsoft Research, Redmond, Washington.
Mentor: Krishna Chintalapudi

Lead Engineer (Research) (July 2017 - July 2019)
Samsung R&D Institute India, Bangalore.

Teaching Experience

Teaching Assistant at USC
Course: CS 551/651: Advanced Computer Networks, Spring 2022, Instructor: Prof. Ramesh Govindan

Guest Lecturer at Princeton University, Topic: Volumetric Video Streaming [PPT]
Course: COS 598a: Machine Learning-Driven Video Systems, Spring 2022, Instructor: Prof. Ravi Netravali


  • Graduate Student Annenberg Fellowship, 2019 - 2023.
  • Received Motorola Gold Medal for best Masters Student, 2018.


  1. SoCC
    Scrooge: A Cost-Effective Deep Learning Inference System
    Hu, Yitao, Ghosh, Rajrup, and Govindan, Ramesh
    In Proceedings of the ACM Symposium on Cloud Computing 2021

    Advances in deep learning (DL) have prompted the development of cloud-hosted DL-based media applications that process video and audio streams in real-time. Such applications must satisfy throughput and latency objectives and adapt to novel types of dynamics, while incurring minimal cost. Scrooge, a system that provides media applications as a service, achieves these objectives by packing computations efficiently into GPU-equipped cloud VMs, using an optimization formulation to find the lowest cost VM allocations that meet the performance objectives, and rapidly reacting to variations in input complexity (e.g., changes in participants in a video). Experiments show that Scrooge can save serving cost by 16-32% (which translate to tens of thousands of dollars per year) relative to the state-of-the-art while achieving latency objectives for over 98% under dynamic workloads.

  2. IoTDI
    Rim: Offloading Inference to the Edge
    Hu, Yitao, Pang, Weiwu, Liu, Xiaochen, Ghosh, Rajrup, Ko, Bongjun, Lee, Wei-Han, and Govindan, Ramesh
    In Proceedings of the International Conference on Internet-of-Things Design and Implementation 2021

    Video cameras are among the most ubiquitous sensors in the Internet-of-Things. Video and audio applications, such as cross-camera activity detection, avatar extraction or language translation will, in the future, offload processing to an edge cluster of GPUs. Rim is a management system for such clusters that satisfies throughput and latency requirements of these applications, while enabling high cluster utilization. It uses coarse-grained knowledge of application structure to profile throughput of applications on resources, then uses these profiles to place applications on cluster nodes to achieve these goals. It dynamically adapts placement to load and failures. Experiments show that on maximal workloads on a testbed, Rim can satisfy requirements of all applications, but competing approaches designed for low-latency GPU execution cannot.

  3. TCPS
    Distributed Scheduling of Event Analytics across Edge and Cloud
    Ghosh, Rajrup, and Simmhan, Yogesh
    ACM Trans. Cyber-Phys. Syst. Jul 2018

    Internet of Things (IoT) domains generate large volumes of high-velocity event streams from sensors, which need to be analyzed with low latency to drive decisions. Complex Event Processing (CEP) is a Big Data technique to enable such analytics and is traditionally performed on Cloud Virtual Machines (VM). Leveraging captive IoT edge resources in combination with Cloud VMs can offer better performance, flexibility, and monetary costs for CEP. Here, we formulate an optimization problem for energy-aware placement of CEP queries, composed as an analytics dataflow, across a collection of edge and Cloud resources, with the goal of minimizing the end-to-end latency for the dataflow. We propose a Genetic Algorithm (GA) meta-heuristic to solve this problem and compare it against a brute-force optimal algorithm (BF). We perform detailed real-world benchmarks on the compute, network, and energy capacity of edge and Cloud resources. These results are used to define a realistic and comprehensive simulation study that validates the BF and GA solutions for 45 diverse CEP dataflows, LAN and WAN setup, and different edge resource availability. We compare the GA and BF solutions against random and Cloud-only baselines for different configurations for a total of 1,764 simulation runs. Our study shows that GA is within 97% of the optimal BF solution that takes hours, maps dataflows with 4–50 queries in 1–26s, and only fails to offer a feasible solution ≤20% of the time.

    Adaptive Energy-Aware Scheduling of Dynamic Event Analytics Across Edge and Cloud Resources
    Ghosh, Rajrup, Komma, Siva Prakash Reddy, and Simmhan, Yogesh
    In 18th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGRID) May 2018

    The growing deployment of sensors as part of Internet of Things (IoT) is generating thousands of event streams. Complex Event Processing (CEP) queries offer a useful paradigm for rapid decision-making over such data sources. While often centralized in the Cloud, the deployment of capable edge devices on the field motivates the need for cooperative event analytics that span Edge and Cloud computing. Here, we identify a novel problem of query placement on edge and Cloud resources for dynamically arriving and departing analytic dataflows. We define this as an optimization problem to minimize the total makespan for all event analytics, while meeting energy and compute constraints of the resources. We propose 4 adaptive heuristics and 3 rebalancing strategies for such dynamic dataflows, and validate them using detailed simulations for 100 - 1000 edge devices and VMs. The results show that our heuristics offer O(seconds) planning time, give a valid and high quality solution in all cases, and reduce the number of query migrations. Furthermore, rebalance strategies when applied in these heuristics have significantly reduced the makespan by around 20 - 25%.

  5. ICAPR
    Exploring the self similar properties for monitoring of air quality information
    Ghosh, Rajrup, Ghosh, Dipanjan, Roy, Sreemoyee, and Mukherjee, Abhik
    In 2015 Eighth International Conference on Advances in Pattern Recognition (ICAPR) Jan 2015

    Air quality information has assumed much importance over the years due to the increase in air pollution. One major hindrance in monitoring of air pollutants is the dearth of spatial availability of aerosol concentration measurements due to the cost involved in deployment of sensors. In this respect, self similarity analysis of data can be very useful. This work is based on standard grid based pollutant dispersion models in a simulated environment over different scales of grid size. The fractal dimension is considered as a scale invariant metric which gives an idea about the variation in pollutant concentration across different scales. A method is detailed for measuring the fractal dimension properties. Results indicate that it is possible to apply the dispersion models across different scales and also the air quality monitored in one region can be compared with other regions.