Title: How do checkpoint mechanisms and power infrastructure failures impact on cloud applications?

Authors: Guto Leoni Santos; Demis Gomes; Djamel Sadok; Judith Kelner; Élisson Rocha; Patricia Takako Endo

Addresses: Federal University of Pernambuco (UFPE), Recife, Pernambuco, Brazil ' Federal University of Pernambuco (UFPE), Recife, Pernambuco, Brazil ' Federal University of Pernambuco (UFPE), Recife, Pernambuco, Brazil ' Federal University of Pernambuco (UFPE), Recife, Pernambuco, Brazil ' Universidade de Pernambuco (UPE), Recife, Pernambuco, Brazil ' Universidade de Pernambuco (UPE), Recife, Pernambuco, Brazil

Abstract: With the growth of cloud computing usage by commercial companies, service providers are looking for ways to improve the quality of their services. Failures in the power subsystem represent a major risk of cloud data centre unavailability at the physical level. At same time, software-level mechanisms (such as application checkpointing) can be used to maintain the application consistency after a downtime and also improve its availability. This paper analyses the impact of power subsystem failures on cloud applications availability, as well as the impact of having checkpoint mechanisms to recover the system from software-level failure. We propose stochastic models to represent the cloud power subsystem, the cloud application, and also the checkpoint-based retrieval mechanisms. The results show that the selection of a given type of checkpoint mechanism does not present a significant impact on the observed metrics. On the other hand, improving the power infrastructure implies gains in performance and availability.

Keywords: cloud data centre; checkpoint mechanisms; availability; performance; stochastic models; power subsystem; stochastic Petri nets; reliability block diagrams; hardware failures; sensitivity analysis; queue system.

DOI: 10.1504/IJGUC.2020.110045

International Journal of Grid and Utility Computing, 2020 Vol.11 No.5, pp.615 - 628

Received: 21 Mar 2018
Accepted: 27 May 2018

Published online: 01 Aug 2020 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article