Synapse

Monthly Newsletter from CNI

Issue - June 2024

The fourth annual CNI Summer School will take place from June 10 to June 14.  Prof. Jean-Francois Chamberland from Texas A&M University will conduct a course on Algorithmic Structures for Emerging Wireless Networks and Statistical Inference in Large Dimensional Spaces. Each day will include two theory sessions from 9:30 AM to 10:30 AM and 10:45 AM to 11:45 AM, followed by a practical session from 12:15 PM to 1:15 PM (Bring your laptops with you!!). The sessions will be held in person at MP-20, Microelectronics & Photonics building, ECE Department, IISc. Don't miss out on this incredible learning opportunity.

Two Papers Accepted at IEEE INFOCOM 2024, including Best Paper Award Winner !

CNI Project Associate, Nithish K Gnani, presented a paper titled "EdgeP4: In-Network Edge Intelligence for a Tactile Cyber-Physical System Testbed Across Cities" at the IEEE International Conference on Computer Communications ( IEEE INFOCOM) , held from May 20-23, 2024, in Vancouver, Canada. Other authors to the paper were former CNI PhD fellow Joydeep Pal, CNI Project Assistant Deepak Choudhary L , and CNI faculty members Dr. T. V. Prabhakar and Dr. Chandramani Singh. It received the CNERT workshop's best paper award. We, at CNI, extend our congratulations to all the authors!

In their work, they focus on improving performance of telesurgery by reducing latency and packet loss. They achieve this by moving the compute of critical operations to the edge devices. The above figure depicts the network architecture designed to achieve this task. Critical operations like pose correction, fiducial marking, and tremor reduction need to be automated to avoid network dependency. Two edge intelligence algorithms were designed and hosted on P4 programmable end switches placed 400 km apart to address this issue. These algorithms compute and command corrective signals locally, implemented entirely on the data plane using Netronome Agilio SmartNICs. The "pose correction" algorithm, placed at the edge switch connected to an industrial robot, achieves a round-trip time of less than 100μs. The "tremor suppression" algorithm, located at the edge switch connected to the operator, reduces physiological tremors, enhancing performance and reducing network load by up to 99.9%. The EdgeP4 framework allows seamless switching between these algorithms based on the tasks performed by connected devices.

Additionally, at the same conference, Joydeep Pal presented another paper titled "Towards a TSN-DetNet Intercity Testbed for Tactile Cyber-Physical Systems" . Other authors to the paper were Deepak Choudhary L, Nithish K Gnani, Dr. T. V. Prabhakar, and Dr. Chandramani Singh.

As shown in the above figure, the authors designed a TSN-DetNet testbed for Tactile Cyber-Physical Systems which enables low-latency and high-reliability communication across two cities 400 kilometers apart. It features two key domains: the CPS domain, which includes haptic devices, robotic arms, and sensing equipment, and the Tactile Internet domain, which uses TSN-enabled LAN switches and MPLS-based WAN routers. This setup accommodates diverse traffic classes, such as Scheduled Traffic for critical tactile data and Best Effort for non-critical flows. Time synchronization is achieved using the Precision Time Protocol . The paper evaluates the testbed's performance metrics, including latency, jitter, packet loss, and time synchronization accuracy. A real-world demonstration of a robotic teleoperation application, with haptic and video feedback, highlights the testbed's practical capabilities. The paper also addresses the challenges of building an intercity TSN-DetNet testbed and outlines future work for demonstrating and evaluating TCPS applications.

Need for Sovereign compute

In this talk, Prof. Manu Awasthi discussed the pervasive presence of digital devices in daily life and the extensive use of computing power in services like Uber, Amazon, and online banking. He explored what "compute" entails and the implications of relying on foreign-designed and manufactured computing infrastructure for India. He also introduced the idea of "Sovereign Compute," highlighting the need to develop and maintain indigenous capabilities in designing, manufacturing, and assembling key components of the computing stack.

Guaranteed performance analysis of networks with bandwidth-sharing schedulers

In this talk, Dr. Anne Bouillard discussed the Deficit Round-Robin (DRR) scheduling algorithm, used in real-time systems and communication networks to allocate bandwidth proportionally to traffic classes. She explained the two-step analysis process: single-node analysis for guaranteed service and per-class analysis for computing delay bounds. She also highlighted the interdependence of these steps in DRR networks with cyclic dependencies and introduced an iterative scheme that improves performance and stability compared to current methods.

Towards Optimizing Storage Costs on the Cloud

In this talk, Koyel Mukherjee discussed optimizing cloud storage and access costs while maintaining performance benchmarks. She explored techniques such as tier optimization and real-time compression performance prediction. Additionally, she introduced access pattern-aware data partitioning as an innovative approach to enhancing efficiency in data management, contrasting it with traditional batch-based methods. Theoretical and empirical studies showcased significant cost savings over platform defaults and closest baselines in literature, validating the effectiveness of the proposed strategies.

Discrimination, A One bit sampling perspective

In this talk, Prof. Sibi Raj Pillai explored the cost-performance tradeoffs in ADCs, especially in low-end devices where economic constraints requires one-bit quantization of input samples. Conventional wisdom suggests the need for large oversampling to compensate for precision limitations in frequency and phase discrimination. However, he shows that this is not necessary by establishing a connection between frequency offset and accumulated one-bit samples post-mixing through Fourier analysis.