ATOM: model-driven autoscaling for microservices

File Description SizeFormat 
ICDCS_ATOM.pdfFile embargoed until 01 January 10000452.77 kBAdobe PDF    Request a copy
Title: ATOM: model-driven autoscaling for microservices
Authors: Gias, A
Casale, G
Woodside, M
Item Type: Conference Paper
Abstract: Microservices based architectures are increasinglywidespread in the cloud software industry. Still, there is ashortage of auto-scaling methods designed to leverage the uniquefeatures of these architectures, such as the ability to indepen-dently scale a subset of microservices, as well as the ease ofmonitoring their state and reciprocal calls.We propose to address this shortage with ATOM, a model-driven autoscaling controller for microservices. ATOM instanti-ates and solves at run-time a layered queueing network model ofthe application. Computational optimization is used to dynami-cally control the number of replicas for each microservice and itsassociated container CPU share, overall achieving a fine-grainedcontrol of the application capacity at run-time.Experimental results indicate that for heavy workloads ATOMoffers around 30%-37% higher throughput than baseline model-agnostic controllers based on simple static rules. We also find thatmodel-driven reasoning reduces the number of actions needed toscale the system as it reduces the number of bottleneck shiftsthat we observe with model-agnostic controllers.
Issue Date: 7-Jul-2019
Date of Acceptance: 29-Mar-2019
URI: http://hdl.handle.net/10044/1/69963
Publisher: IEEE
Copyright Statement: This paper is embargoed until publication.
Sponsor/Funder: European Commission
Funder's Grant Number: 825040
Conference Name: IEEE International Conference on Distributed Computing Systems (ICDCS)
Publication Status: Accepted
Start Date: 2019-07-07
Finish Date: 2019-07-10
Conference Place: Dallas, Texas, USA
Embargo Date: publication subject to indefinite embargo
Appears in Collections:Computing



Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

Creative Commons