Resource management is traditionally addressed by policies implemented inside the resource provider. Here we study the problem with an attitude that is specular but complementary, which consists in designing a distributed client-side access regulation algorithm that improves the utilization of an elastic resource. The introduction of elastic resources — a feature of the cloud computing paradigm — complicates their management since, when the workload applied on the resource varies (for instance with the number of users) the resource automatically follows such variations with its capacity. But the presence of an extra computational cost related with capacity variations motivates a non linear, lazy response, that penalizes dynamic environments. Hence the interest for an algorithm that shapes the production of service requests on the client side. To make our investigation more adherent to a practical environment, we introduce a real time requirement: each client must have access to the service at least every π time units. Examples of this requirement, that features a bounded degree of asynchrony, are found, for instance, in network streaming applications: stream chunks must feed the input buffer at the destination. The algorithm we investigate is based on the random walk of a token. To evaluate the range of applicability of the algorithm, we define an analytic model of its stochastic behavior — described by a non-Markov process — and then we compare its performance with a benchmark algorithm, representative of an effective solution that is often used in practice.

Improving the Utilization of an Elastic Resource: A Client-side Approach

CIUFFOLETTI, AUGUSTO
2013

Abstract

Resource management is traditionally addressed by policies implemented inside the resource provider. Here we study the problem with an attitude that is specular but complementary, which consists in designing a distributed client-side access regulation algorithm that improves the utilization of an elastic resource. The introduction of elastic resources — a feature of the cloud computing paradigm — complicates their management since, when the workload applied on the resource varies (for instance with the number of users) the resource automatically follows such variations with its capacity. But the presence of an extra computational cost related with capacity variations motivates a non linear, lazy response, that penalizes dynamic environments. Hence the interest for an algorithm that shapes the production of service requests on the client side. To make our investigation more adherent to a practical environment, we introduce a real time requirement: each client must have access to the service at least every π time units. Examples of this requirement, that features a bounded degree of asynchrony, are found, for instance, in network streaming applications: stream chunks must feed the input buffer at the destination. The algorithm we investigate is based on the random walk of a token. To evaluate the range of applicability of the algorithm, we define an analytic model of its stochastic behavior — described by a non-Markov process — and then we compare its performance with a benchmark algorithm, representative of an effective solution that is often used in practice.
Ciuffoletti, Augusto
File in questo prodotto:
File Dimensione Formato  
paper.pdf

accesso aperto

Tipologia: Documento in Post-print
Licenza: Tutti i diritti riservati (All rights reserved)
Dimensione 224.14 kB
Formato Adobe PDF
224.14 kB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11568/226526
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact