Research

Design of WebRTC based Real Time Streaming System


With the start of pandemic, there was an increase in the demand of video conferencing systems. These systems were highly scalable but they were always criticised for their lack of data security. This project aimed at creating a video conferencing system with a distributed architecture, keeping data security at the helm. I developed a web application with a mesh architecture using WebRTC. I also added a discovery server, capability to view multiple screens and selective display. This design could support 5 users with video but wasn’t scalable beyond that. I used supernodes to solve this problem. These improved the scalability of the system to 12 users.

Cluster Middleware: Fenrir


An cluster middleware for resource management with a master-worker architecture. The master node accepts tasks along with their requirements such as cpu and memory. Then, It identifies the appropriate worker node from the available pool and allocates the task. Some features of this system are:

  • The master node has a backup node and works in an active-passive mode.
  • The worker nodes continuously send heartbeats to the master node for failure detection. Upon failure of a node, The master node reassigns any tasks on that node to a separate node.
  • The system is priority aware and uses dynamic priority scheduling of tasks to prevent job starvation

Efficient task scheduling for autonomous driving DNN workloads


This work is an implementation of the paper S3DNN.
DNN workloads for autonomous driving typically consists of relatively reduced load near the last layers. This work leverages this to improve the efficiency of resource usage.

Performance Behavior Analysis and Enhancement Of System Calls


File Systems for Distributed Storage