BLESSON VARGHESE
School of Computer Science, University of St Andrews
SOFTWARE
'Working Code Trumps All Hype'
Courtesy: Professor M. Satyanarayan, Carnegie Mellon University
SCISSIONTL
ScissionTL is a tool for automated benchmarking of distributed DNNs on a given set of target device, edge and cloud resources while minimising the data transfer using transfer layers for maximising the performance of DNN inference.
SCISSION
Scission is a tool for automated benchmarking of distributed DNNs on a given set of target device, edge and cloud resources for identifying the execution approach and determining the optimal partition for maximising the performance of DNN inference.
DEEP NEURAL NETWORK ADAPTIVITY
The software aims to identify whether there is a case for adaptivity - the need for repartitioning DNNs to adapt to changing operational conditions (CPU, memory, or network) at the network edge once it is already deployed across the cloud and the edge.
DEFOG
DeFog is an open source Fog/Edge benchmarking suite comprising six benchmarks: (1) YOLO - deep learning-based object recognition, (2) PocketSphinx - text-to-speech converter, (3) Aeneas - text-to-audio forced alignment, (4) iPokeMon - geolocation based mobile game, (5) FogLAMP - IoT edge gateway application, and (6) RealFD - real-time face detection on video streams
ENORM
ENORM is a framework for Edge NOde Resource Management. The framework realises an architecture and a few fundamental algorithms for managing Edge resources. The framework is demonstrated on a PokeMon Go-like online game.
DYVERSE
DYVERSE (DYnamic VERtical Scaling in Multi-tenant Edge Environments) builds on ENORM while providing priority-based dynamic resource allocation for supporting multiple Edge applications on an individual Edge node. The research is demonstrated on multiple applications.
EDGE FAIR SCHEDULER
The Edge is anticipated to be resource constrained and a competitive environment to acquire resources for services offloaded (jobs) from the Cloud (client) to the Edge. Edge Fair Scheduler (EFS) is a prototype for fairly distributing jobs on an individual Edge node while being fair to multiple clients and the priorities associated with the jobs.