Back to search

Boston University

Orchestrating cloud resources to optimize performance and cost

Abstract

dc:description.abstract

In the last decade, Function-as-a-Service (FaaS) became one of the popular choices for building and deploying cloud applications. Compared to Infrastructure-as-a-Service (IaaS), FaaS offers an abstraction of backend management, an easy programming model, low cold starts, and a true “pay as you go” pricing model. While efficient and relatively simpler, the cost and performance of an application, when deployed using FaaS, can be adversely affected if not properly managed and configured. Previous approaches have advocated the limited use of FaaS while scaling out the virtual machine (VM) based resources to avoid Service-Level-Objective (SLO) violations. However, these approaches miss out on potential long-term cost savings by employing FaaS consistently. Similarly, to manage a FaaS deployment, various machine learning and optimization techniques have been suggested but these techniques either have high costs or fall short as they fail to adapt to the dynamic nature of FaaS platforms. To this end, we present Thrifty, a hybrid approach to leveraging FaaS in conjunction with other cloud services to optimize both cost and performance. Thrifty consists of two main components: 1) LIBRA: a load-balancing framework to utilize IaaS and FaaS resources efficiently. Based on the demand, it decides to use either FaaS, IaaS, or both to maximize cost savings while meeting the SLOs; 2) xCOSE: a resource configuration and placement technique for FaaS deployments. It addresses the performance variability of FaaS platforms and meets the SLO by adapting the resource configurations with minimal sampling cost. It can configure single- and multi-function (service graph) applications. We evaluate Thrifty in extensive simulations and on the Amazon Web Services (AWS) cloud platform using real applications. Our evaluations show that consistent and opportunistic usage of FaaS through LIBRA can reduce SLO violations by up to 85% and cost by up to 53% when compared to other approaches to deploying cloud applications. Furthermore, xCOSE has the ability to configure simple and complex FaaS applications with minimal sampling cost.

Degree

thesis:*
Name thesis:degree_name
Doctor of Philosophy
Level thesis:degree_level
doctoral
Discipline thesis:degree_discipline
Computer Science
Grantor
Boston University
Year dc:date.issued
2023

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Raza, Ali
Advisor dc:contributor.advisor
  • Matta, Abraham
ORCID dc:identifier.orcid

Subjects

dc:subject × 7

Rights

dc:rights
Statement dc:rights
  • Attribution-NoDerivatives 4.0 International
Language dc:language.iso
en_US

Identifiers

dc:identifier.*
Handle dc:identifier.uri
https://hdl.handle.net/2144/49670

Chain of custody

source
Harvested from
Boston University
Base URL
open.bu.edu/oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Raza, Ali. Orchestrating cloud resources to optimize performance and cost. doctoral thesis, Boston University, 2023. https://hdl.handle.net/2144/49670