Fault Tolerant Horizontal Computation Offloading
Droob, Alexander; Morratz, Daniel; Jakobsen, Frederik Langkilde; Carstensen, Jacob; Mathiesen, Magnus; Bohnstedt, Rune; Albano, Michele; Moreschini, Sergio; Taibi, Davide (2023)
Droob, Alexander
Morratz, Daniel
Jakobsen, Frederik Langkilde
Carstensen, Jacob
Mathiesen, Magnus
Bohnstedt, Rune
Albano, Michele
Moreschini, Sergio
Taibi, Davide
Teoksen toimittaja(t)
Ardagna, Claudio
Awaysheh, Feras
Bian, Hongyi
Chang, Carl K.
Chang, Rong N.
Delicato, Flavia
Desai, Nirmit
Fan, Jing
Fox, Geoffrey C.
Goscinski, Andrzej
Jin, Zhi
Kobusinska, Anna
Rana, Omer
IEEE
2023
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-2023120510477
https://urn.fi/URN:NBN:fi:tuni-2023120510477
Kuvaus
Peer reviewed
Tiivistelmä
The broad development and usage of edge devices has highlighted the importance of creating resilient and computationally advanced edge-to-cloud continuum environments. When working with edge devices these desiderata are usually achieved through replication and offloading. This paper reports on the design and implementation of a fault-tolerant service that enables the offloading of jobs from devices with limited computational power. We propose a solution that allows users to upload jobs through a web service, which will be executed on edge nodes within the system. The solution is designed to be fault tolerant and scalable, with no single point of failure as well as the ability to accommodate growth, if the service is expanded. The use of Docker checkpointing on the worker machines ensures that jobs can be resumed in the event of a fault. We provide a mathematical approach to optimize the number of checkpoints that are created along a computation, given that we can forecast the time needed to execute a job. We present experiments that indicate in which scenarios checkpointing benefits job execution. Our experiments shows the benefits of using checkpointing and restore when the completion jobs' time rises compared with the forecast fault rate.
Kokoelmat
- TUNICRIS-julkaisut [19020]