Publications

PowerTrain: Fast, Generalizable Time and Power Prediction Models to Optimize DNN Training on Accelerated Edges

Prashanthi S. K., Saisamarth Taluri, Beautlin S, Lakshya Karwa and Yogesh Simmhan, Elsevier Future Generation Computer Systems (FGCS) 2024

Performance Characterization of Containerized DNN Training and Inference on Edge Accelerators

Prashanthi S.K., Vinayaka Hegde, Keerthana Patchava, Ankita Das and Yogesh Simmhan, IEEE International Conference on High-Performance Computing, Data and Analytics (HiPC) 2023

Characterizing the Performance of Accelerated Jetson Edge Devices for Training Deep Learning Models

Prashanthi S.K., Sai Anuroop Kesanapalli and Yogesh Simmhan, International Conference on Measurement and Modeling of Computer Systems (SIGMETRICS Abstracts 2023)

Characterizing the Performance of Accelerated Jetson Edge Devices for Training Deep Learning Models

Prashanthi S.K., Sai Anuroop Kesanapalli and Yogesh Simmhan, Proceedings of the ACM on Measurement and Analysis of Computing Systems (POMACS 2022)

Towards efficient scheduling of concurrent DNN training and inferencing on accelerated edge devices

Prashanthi S K, Vinayaka Hegde and Yogesh Simmhan, Accepted at EuroSys Doctoral workshop (EuroDW), European Conference on Computer Systems (EuroSys) 2023

Towards efficient scheduling of concurrent DNN training and inferencing on accelerated edge devices

Prashanthi S K, Vinayaka Hegde and Yogesh Simmhan, Accepted at Student Showcase, IEEE/ACM International Symposium on Cluster, Cloud and Internet Computing (CCGRID) 2023

Don't Miss the Train: A Case for Systems Research into Training on the Edge

Prashanthi S.K, Aakash Khochare, Sai Anuroop Kesanapalli, Rahul Bhope, and Yogesh Simmhan
Workshop on Parallel AI and Systems for the Edge (PAISE 2022) at 36th International Parallel and Distributed Processing Symposium (IPDPS 2022)

Characterizing the Performance of Deep Learning Workloads on Accelerated Edge Computing Devices

Prashanthi S.K, Sai Anuroop Kesanapalli, Aakash Khochare and Yogesh Simmhan
IEEE International Conference on High Performance Computing, Data & Analytics (HiPC) Student Research Symposium, 2021

Design and analysis of a minimally invasive and ECG controlled Ventricular Assistive Device

Prajwal Sharma, Prashanthi K, Krishna Nagaraj, Vinay Chandraekhar, Vikas Vazhiyal, and Madhav Rao
32nd International Conference on VLSI Design and 2019 18th International Conference on Embedded Systems (VLSID), 2019

Development of a minimally invasive Ventricular Assistive Device

Prajwal Sharma, Prashanthi K, Krishna Nagaraj, Vikas Vazhiyal, and Madhav Rao
IEEE Sensors 2018

A Quantitative Analysis of the Real-time Capabilities of Linux with PREEMPT_RT

Prashanthi K, Deepika Raj, B Thangaraju
Open Source For You, April 2018

Research/Academic Projects

Characterizing the Performance of Accelerated Edge Devices for Training Deep Learning Models

Studied the impact of various training parameters on resource utilization, training time and energy, identified bottlenecks and resource inter-dependencies

An exploration of system virtualization on the edge

Carried out detailed CPU, memory, network and disk benchmarking on single and co-located KVM and Firecracker virtual instances on Nvidia Jetson NX to understand the overheads of virtualization

Elfstore: A Resilient Data Storage Service for Federated Edge and Fog Resources

Worked on simulating state transition diagrams and evaluating them on various performance metrics to ensure optimal utilization of storage while guaranteeing a given reliability

A comprehensive assistive system for the visually impaired

Color and pattern recognition of fabrics, OCR on menu and visiting cards, face detection with voice input and output implemented on a Pi intended to be used as a wearable system.

Device Driver for Controlling a Projector using a Raspberry Pi

Kernel modules were created to detect the HDMI unplug uevent and transmit the hexcodes for powering off the projector connected to the Pi.

Remote Warehouse Monitoring

The iRobot moves autonomously collecting information from various sensors which is then periodically transmitted to a server and thereafter the cloud for further monitoring.

Expression Compiler for ARM

Developed a compiler that reads input from a file and generates assembly instructions for ARM Cortex M4. Handled arithmetic, logical, relational and shift operators, assignment statements, if-then-else block.

Implementation of a real-time VoIP phone

The client and server sides of a VoIP phone were implemented. Used PulseAudio to pick up voice packets and G711 to encode. A periodic scheduler was integrated with real-time protocol to ensure good response time.

Smart object finder using Bluetooth

A small Bluetooth tag is attached to the objects of interest and an android app on the user’s phone is used to locate the object in an indoor environment.

Industry Projects/Responsibilities

GPU Scheduler Performance Analysis and Tuning

Identified and executed several PoC experiments to characterise the performance of scheduler.
Identified and fixed gaps that led to performance improvements of 5-20% on Gen9+ platforms across various benchmarks, and was awarded for the same.

Telemetry Lead, Graphics Kernel Mode Driver (Gen 9)

Resolved Live Kernel Events (BSODs,TDRs) using kernel crash dumps received as part of telemetry data.
Added instrumentation that could help in effective debug without reproduction steps.

Kernel Mode Driver Lead, Revenue Products

Contributed to the development and testing of several features in the Driver and Scheduler firmware.
Debugged and resolved critical launch-blocking OEM bugs in Media firmware and Graphics scheduler.