Publication Details
A Reality Check on Inference at Mobile Networks Edge
Kocour Martin, Ing. (DCGM)
RAMAN, A.
LEONTIADIS, I.
Luque Jordi
SASTRY, N.
NUNEZ-MARTINEZ, L.
PERINO, D.
PERALES, C.
Edge computing, Artificial Intelligence
Edge computingis considered a key enabler to deploy ArtificialIntelligenceplatforms to provide real-time applications such asAR/VR or cognitive assistance. Previousworks show computingcapabilitiesdeployed very close to the user can actually reduce theend-to-end latency of such interactiveapplications. Nonetheless,themain performance bottleneck remains in the machine learninginference operation. In this paper, wequestion some assumptionsofthese works, as the network location where edge computing isdeployed, and considered softwarearchitectures within the frame-workof a couple of popular machine learning tasks. Our experimen-tal evaluation shows that after performancetuning that leveragesrecentadvances in deep learning algorithms and hardware, net-work latency is now the main bottleneck onend-to-end applicationperformance.We also report that deploying computing capabilitiesat the first network node still provideslatency reduction but, over-all,it is not required by all applications. Based on our findings, weoverview the requirements and sketch thedesign of an adaptivearchitecturefor general machine learning inference across edgelocations.
Edge computing is considered a key enabler to deploy Artificial Intelligence platforms to provide real-time applications such as AR/VR or cognitive assistance. Previous works show computing capabilities deployed very close to the user can actually reduce the end-to-end latency of such interactive applications. Nonetheless, the main performance bottleneck remains in the machine learning inference operation. In this paper, we question some assumptions of these works, as the network location where edge computing is deployed, and considered software architectures within the framework of a couple of popular machine learning tasks. Our experimental evaluation shows that after performance tuning that leverages recent advances in deep learning algorithms and hardware, network latency is now the main bottleneck on end-to-end application performance. We also report that deploying computing capabilities at the first network node still provides latency reduction but, overall, it is not required by all applications. Based on our findings, we overview the requirements and sketch the design of an adaptive architecture for general machine learning inference across edge locations.
@inproceedings{BUT156850,
author="CARTAS, A. and KOCOUR, M. and RAMAN, A. and LEONTIADIS, I. and LUQUE, J. and SASTRY, N. and NUNEZ-MARTINEZ, L. and PERINO, D. and PERALES, C.",
title="A Reality Check on Inference at Mobile Networks Edge",
booktitle="Proceedings of the 2nd ACM International Workshop on Edge Systems, Analytics and Networking (EDGESYS '19)",
year="2019",
pages="54--59",
publisher="Association for Computing Machinery",
address="Dressden",
doi="10.1145/3301418.3313946",
isbn="978-1-4503-6275-7",
url="https://dl.acm.org/citation.cfm?doid=3301418.3313946"
}