Synapse

Monthly Newsletter from CNI

Issue - November 2024

Join us for the upcoming talk on Minimizing the impact of Bufferbloat by Prof. Mohit Tahiliani from NIT Surathkal on 12/11/2024.

Join us for the upcoming talk on "Adaptive discretization-based reinforcement learning for average reward Markov decision processes" by Avik Kar from IISc on 19/11/2024

Join us for the upcoming talk on Interventional Causal Representation Learning by Prof. Tajer Ali, RPI on 20/11/2024.

Join us for the upcoming seminar on 26/11/2024 with Prof. Lalitha Vadlamani from IIT Hyderabad.

DoT Workshop on Bharat 6G Vision: Bridging Research and Standards for India's 6G Future

The "Bharat 6G Vision: Translating Research to Standards" workshop, held on September 2-3, 2024, at the Indian Institute of Science, Bangalore, gathered influential leaders from academia, industry, and government to discuss and shape the future of 6G standards in India. Inaugurated by Dr. Neeraj Mittal, Secretary of Telecom, DoT HQ, the event provided a platform for key insights and strategic discussions aimed at advancing India’s role in the global 6G landscape. On the first day, Dr. Kumar N. Sivarajan (CTO, Tejas Networks), Prof. Rajesh Sundaresan (Dean, EECS, IISc), Prof. Pradipta Biswas (Associate Professor, CPDM, IISc), and Dr. Pamela Kumar (Chief Strategy Advisor, FSID) shared perspectives on the current state and future needs of ICT standardization. Mr. Robert Clark, Advisor at ITU HQ Geneva, led sessions on ITU-T standards, explaining processes for contribution, approval, and effective participation in the global telecom ecosystem.

The second day shifted focus to intellectual property rights and patents, essential to fostering innovation in ICT. Prof. Arul George Scaria from NLSIU delivered a detailed presentation on patents, copyrights, and trademarks in the context of ICT, followed by a panel discussion on standard essential patents (SEPs) with Sh. Santosh Gupta (Assistant Controller of Patents, DPIIT), Sh. Kandasamy S. (VP, Signalchip), and Dr. Vijay Singh (Manager, IPTel, IISc), who discussed policy, licensing, and support mechanisms for SMEs. The workshop concluded with a session on the upcoming World Telecom Standardization Assembly (WTSA) 2024, presented by Sh. R. Shakya and Ms. Akanksha Sharma from the ITU Area Office, providing insights into India’s preparations and contributions for the global 6G standards framework.

EdgeRIC: Empowering AI-based Real-Time Intelligent Control, Optimization and Security in NextG Networks

In this talk, Prof. Dinesh Bharadia introduced EdgeRIC, an advanced real-time RAN intelligent controller designed to meet the demands of modern applications like robotics, AR/VR, and automotive systems, which traditional network optimization techniques struggle to support. Their work leverages reinforcement learning, to improve ORAN (Open RAN) stack performance. Their results show that EdgeRIC, with a digital twin for wireless channel consistency, significantly enhances user experience, achieving up to a 90% reduction in video streaming stalls. Prof. Bharadia also discussed innovative MIMO applications that maintain connectivity and cancel interference in urban settings, demonstrating the real-world impact of these advancements on NextG networks.

Load Balancing under Data Locality: Extending Mean-Field Framework to Constrained Large-Scale Systems

In this talk, Prof. Debankur Mukherjee explored how data locality constraints complicate load balancing in large-scale systems like data centers and cloud networks, where traditional mean-field approximations often fall short. He presented recent progress in extending the accuracy of mean-field approximations for constrained systems, which model task-to-server assignments as a bipartite graph. Their results reveal that optimal performance depends significantly on the structural properties of this graph, such as being an expander or a spatial graph. This work advances our understanding of load balancing in complex, constrained environments, with implications for future data center architectures.

Efficiently Serving Large Language Models

In this talk, Ashish Panwar discussed the challenges and solutions for efficiently serving Large Language Models (LLMs), whose deployment is resource-intensive due to high GPU requirements and typically low utilization rates. He highlighted Microsoft's recent developments, such as Sarathi-Serve and vAttention, that address core issues like scheduling and memory management in LLM serving systems. Their work demonstrates that optimizing resource usage in LLM serving setups can drastically reduce operational costs and improve service quality, making these models more accessible for widespread applications.

Understanding Performance of Internet Video using Network Measurement Data 

In this talk, Prof. Tarun Mangla addressed the growing need for last-mile ISPs to manage networks effectively in response to increasing demand for video streaming. He presented eMIMIC, a method that estimates video Quality of Experience (QoE) by analyzing network traffic patterns in encrypted video streams, bypassing the need for endpoint data. Additionally, he discussed a machine-learning approach for QoE metrics in video conferencing, effective without access to application-layer headers. These techniques enable ISPs to assess and improve QoE with minimal access to private data, providing valuable insights for network optimization in video applications.