DREAM:Lab Team wins the IEEE TCSC SCALE Challenge for 2019

The DREAM:Lab team, led by PhD candidate Aakash Khochare, won the IEEE TCSC SCALE Challenge 2019 award [1]. The annual competition sponsored by the IEEE Technical Committee on Scalable Computing (TCSC) [2] and held as part of the IEEE/ACM Cluster, Cloud and Grid Computing Conference (CCGrid) [3] recognizes “real-world problem solving using computing that scales”. The team also included students Sheshadri Ramachandra, Shriram Ramesh, Swapnil Gandhi and Anubhav Guleria, and the faculty advisor Prof. Yogesh Simmhan, all from the Department of Computational and Data Sciences (CDS), Indian Institute of Science (IISc), Bangalore. The entry from IISc was one of three that were selected for demonstration at the SCALE Challenge this year, with the other two from Tsinghua University in China and University of Sydney in Australia [4]. This is the first time that a team from India has won the SCALE challenge in the 12 years that the competition has been held, and the second time that Prof. Simmhan has won this challenge.

Aakash Khochare and Yogesh Simmhan receiving the SCALE Challenge award from George Pallis, Program Co-Chair of CCGrid 2019 SCALE Challenge 2019 Certificate

The highlight of the competition is a live demo on “end-to-end problem solving using concepts, technologies and architectures that facilitate scaling”.  The winning entry, titled Dynamic Scaling of Video Analytics for Wide-area Tracking in Urban Spaces [5], involved the development and demonstration of a distributed Edge Computing platform, Anveshak, for video inference and classification using Deep Neural Networks over concurrent city-wide camera feeds, and using these to automatically control the traffic signals over a road network to offer a “green wave” for emergency vehicles that are detected by the cameras in real-time. The simulation demonstrated on the Bangalore road network showed that using Anveshak reduced the time taken for an ambulance to move across a 3.5km stretch of the road from 385 seconds to 203 seconds, giving that much more time within the “golden hour” for the patient to be rescued.

 Ambulance routing using default traffic signalling by SUMO   Ambulance routing using Anveshak’s video analytics and traffic signalling

 The demonstration used the SUMO traffic simulator for Bangalore roads, with up to 29,000 road segments totaling 2000kms, 20,000 vehicles in motion, and 600 traffic lights. Up to 4000 cameras were present in the road network, replaying real-world traffic feeds. Anveshak was run on up to 4025 Kubernetes containers on a lab cluster with 36 servers hosted at CDS, which behaved as edge, fog and cloud resources. Existing YOLO and Inception neural network models were retrained and used to detect the ambulance from live feeds. A unique feature of the platform is its a spotlight tracking algorithm that dynamically limits the scope of video streams that need to be processed based on the expected spatial location of the emergency vehicle being tracked, thus requiring it to concurrently process only a small fraction of all the feeds.

slides

This year’s SCALE Challenge was held at the CCGrid conference in Larnaca, Cyprus between 14-17 May, 2019 [1,4]. Besides Anveshak, two other teams were short-listed for the presentation and demonstration. The presentation, “Million-Core-Scalable Simulation of the Elastic Migration Algorithm on Sunway TaihuLight Supercomputer” from Tsinghua University, National Supercomputing Center in Wuxi and Shandong University in China showcased the use of custom CPU architecture and register optimizations to efficiently use 2 million CPU cores on Sunway TaihuLight, the world’s third fastest supercomputer, for geological simulations of oil beds. The University of Sydney and Macquarie University in Australia demonstrated “Scalable Video Transcoding in Public Clouds“, which used thread parallelism and message-queue optimizations for transcoding video segments on a cloud data center. All entries are accompanied by papers that will appear as part of the CCGrid 2019 conference proceedings.

camera-ready

References

[1] IEEE TCSC SCALE Challenge, 2019, https://www.ccgrid2019.org/pages/scale2019.html

[2] IEEE Technical Committee on Scalable Computing (TCSC), https://www.ieeetcsc.org/

[3] IEEE/ACM CCGrid Conference Series, http://www.buyya.com/ccgrid/

[4] IEEE/ACM CCGrid 2019 Program, https://www.ccgrid2019.org/pages/program.html

[5] Dynamic Scaling of Video Analytics for Wide-area Tracking in Urban Spaces, Aakash Khochare, Sheshadri K. R., Shriram R. and Yogesh Simmhan, in Proceedings of the 2019 19th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGRID), 14-17 May 2019, Larnaca Cyprus