Our overarching aim is to research, design and implement operating systems to support emerging applications with requirements that are data and model driven with machine in the loop and are therefore inherently more demanding.
Every year we host an info-session on the various projects being carried out at EPL. You can find the slides and recording for this year here:
Note, you need organizational(GT) access for the above.
Here are some themes that we currently work on:
Federated Orchestration for Edge Computing . With growing interest in serverless computing, we envision situation-awareness applications like autonomous driving to also be developed as FaaS functions to extoll the many benefits of Serverless computing! This includes reducing developer burden and improving resource scheduling at function invocation granularity. This project explores a distributed system for resource orchestration and challenges that may arise in an edge computing scenario where resources are constrained and can easily saturate at a local edge computing node. Federated shared use of resources across multiple such edge nodes can help achieve efficient use of resources while ensuring application service level objectives are met.
Bandwidth Aware Application Management on wireless networks. Wireless mesh networks are used to serve many communities where wired internet infrastructure is hard to deploy/does not exist. In such networks, the community-wide access is available to locally deployed applications like video conferencing servers, forums and street camera systems. Managing these applications, whose performance varies depending on the capacity of these wireless links is difficult with today’s schedulers, which often do not consider link capacity as a resource in the cluster. To make matters worse, wireless links also experience temporal variations in capacity due to temporary obstruction, fading and interference. This project tries to mitigate the impact of bandwidth variations on application performance by taking bandwidth into account during application deployment, and performing application component rescheduling, in the face of variations in link capacity.