event
PhD Defense by Anirudh Sarma
Primary tabs
Title: Performance-aware Resource Conservation Techniques for Geo-distributed Edge Infrastructures
Date: Monday, April 20th, 2026
Time: 11:30 AM - 12:30 PM (Public portion) | 11:30 AM - 1:30 (Full)
Location: Klaus Room 2100 (Teams Meeting Link)
Anirudh Sarma
CS PhD Candidate
Embedded Pervasive Laboratory | School of Computer Science
College of Computing | Georgia Institute of Technology
Committee:
Dr. Umakishore Ramachandran (Advisor) - School of Computer Science, Georgia Institute of Technology
Dr. Alexandros Daglis - School of Computer Science, Georgia Institute of Technology
Dr. Alexey Tumanov - School of Computer Science, Georgia Institute of Technology
Dr. Francisco Romero - School of Computer Science, Georgia Institute of Technology
Dr. Myungjin Lee - Cisco Research
Abstract:
In a future filled with immersive virtual/augmented reality experiences and autonomous vehicles that need to respond to emergent situations, next-generation applications will generate enormous amounts of data and demand latency-sensitive processing to convert information to actionable knowledge. Edge computing has emerged to meet application demands where servers are deployed in geo-distributed sites close to the edge of the network, i.e., the source of information. Due to space and energy constraints, such edge sites have limited resources compared to traditional cloud datacenters.
To efficiently utilize the limited resources available at the edge, this dissertation proposes a granular resource management framework that conserves network, storage, and CPU resources at the Edge. The techniques enshrined in this framework are demonstrated via four main contributions -
1) ClairvoyantEdge, a networked-edge system that conserves network and storage resources; it prefetches static video data on to the edge servers, enables content reuse and delivery from edge servers to the clients via mmWave links thus reducing the burden on the backhaul bandwidth by 50% and consequently conserving the backhaul bandwidth for real-time traffic.
2) HarvestContainers, a system that conserves CPU resources by harvesting spare CPU cores from latency-sensitive containers without impacting their performance and making such cores available for other throughput-oriented applications hosted at the edge site.
3) FEO, a fully decentralized resource orchestrator that offloads computations (expressed as functions in the Function-as-a-Service paradigm) to neighboring edge sites with spare capacity to help meet their service level objectives (SLO) and conserve CPU resources across the geo-distributed edge infrastructure.
4) MISO, in-memory state orchestrator that augments FEO to enable computation offload for stateful functions. Furthermore, it selectively migrates state to improve data locality and eliminate redundant network transfers, improving application performance with fewer compute and network resources.
Groups
Status
- Workflow status: Published
- Created by: Tatianna Richardson
- Created: 04/19/2026
- Modified By: Tatianna Richardson
- Modified: 04/19/2026
Categories
Keywords
User Data
Target Audience