TY - JOUR
T1 - CoxNet
T2 - A Computation Reuse Architecture at the Edge
AU - Bellal, Zouhir
AU - Nour, Boubakr
AU - Mastorakis, Spyridon
N1 - Funding Information:
Manuscript received December 31, 2020; revised February 26, 2021 and March 19, 2021; accepted April 2, 2021. Date of publication April 7, 2021; date of current version May 20, 2021. This work was supported in part by the National Institutes of Health under Grant NIGMS/P20GM109090; in part by the National Science Foundation under Award CNS-2016714; and in part by the Nebraska University Collaboration Initiative. (Corresponding author: Boubakr Nour.) Zouhir Bellal is with the LabRI-SBA Lab, Ecole Superieure en Informatique, Sidi Bel Abbes 2045, Algeria (e-mail: [email protected]).
Publisher Copyright:
© 2017 IEEE.
PY - 2021/6
Y1 - 2021/6
N2 - In recent years, edge computing has emerged as an effective solution to extend cloud computing and satisfy the demand of applications for low latency. However, with today's explosion of innovative applications (e.g., augmented reality, natural language processing, virtual reality), processing services for mobile and smart devices have become computation-intensive, consisting of multiple interconnected computations. This coupled with the need for delay-sensitivity and high quality of service put massive pressure on edge servers. Meanwhile, tasks invoking these services may involve similar inputs that could lead to the same output. In this paper, we present CoxNet, an efficient computation reuse architecture for edge computing. CoxNet enables edge servers to reuse previous computations while scheduling dependent incoming computations. We provide an analytical model for computation reuse joined with dependent task offloading and design a novel computing offloading scheduling scheme. We also evaluate the efficiency and effectiveness of CoxNet via synthetic and real-world datasets. Our results show that CoxNet is able to reduce the task execution time up to 66% based on a synthetic dataset and up to 50% based on a real-world dataset.
AB - In recent years, edge computing has emerged as an effective solution to extend cloud computing and satisfy the demand of applications for low latency. However, with today's explosion of innovative applications (e.g., augmented reality, natural language processing, virtual reality), processing services for mobile and smart devices have become computation-intensive, consisting of multiple interconnected computations. This coupled with the need for delay-sensitivity and high quality of service put massive pressure on edge servers. Meanwhile, tasks invoking these services may involve similar inputs that could lead to the same output. In this paper, we present CoxNet, an efficient computation reuse architecture for edge computing. CoxNet enables edge servers to reuse previous computations while scheduling dependent incoming computations. We provide an analytical model for computation reuse joined with dependent task offloading and design a novel computing offloading scheduling scheme. We also evaluate the efficiency and effectiveness of CoxNet via synthetic and real-world datasets. Our results show that CoxNet is able to reduce the task execution time up to 66% based on a synthetic dataset and up to 50% based on a real-world dataset.
KW - Edge computing
KW - computation reuse
KW - serverless computing
KW - service offloading
UR - http://www.scopus.com/inward/record.url?scp=85103893022&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85103893022&partnerID=8YFLogxK
U2 - 10.1109/TGCN.2021.3071497
DO - 10.1109/TGCN.2021.3071497
M3 - Article
AN - SCOPUS:85103893022
SN - 2473-2400
VL - 5
SP - 765
EP - 777
JO - IEEE Transactions on Green Communications and Networking
JF - IEEE Transactions on Green Communications and Networking
IS - 2
M1 - 9397772
ER -