Dependability analysis for characterizing Google cluster reliability

Abstract

Cloud solutions are emerging as a new suitable way of transforming traditional IT data centers to highly available and reliable computing resources for hosting critical applications and data. However, software and hardware failures are a common problem in cloud datacenters that can lead to harmful damages. In this paper, we analyze the physical server failures in the Google cloud datacenter. We study the Google cluster properties to investigate the relationship among physical servers' failure rate and jobs failure events. The failure rate of Google cluster executed jobs and servers is taken into consideration during a 29-day period. We present a reliability model for Google cluster physical machines using the continuous time Markov chains according to this observation. We attempt to analyze the obtained model through SHARPE software packages to improve the understanding of failure events in the Google cloud cluster. We also explore the cluster availability based on parameters like steady-state availability, steady-state unavailability, mean time to failure, and mean time to repair in the Google cluster.

A survey on correlation and its structural analysis method for failure factors in cloud service

“…Mesbahi MR et al [12,13] finds a medium positive correlation between job failure and machine failure rate and a weak positive correlation between job failure and machine updates in Google clouds. However, this paper only considered the causes of job failure in terms of physical failures and did not consider the influence of their running factors.…”

Causality between violations and failure factors in cloud service and its analysis methods: a survey

“…On the one hand, in [4] [5], They research the statistical characteristics between failed jobs and physical machines in Google Cloud Services. They then analyze and find that job failure has a moderate positive correlation with machine failure and only a weak positive correlation with machine updates.…”

Combining Markov and Semi-Markov Modelling for Assessing Availability and Cybersecurity of Cloud and IoT Systems

“…The analyzed references considering Markov modelling for assessing the availability and cybersecurity of cloud and IoT systems can be divided into three groups: the references related to assessing availability and reliability [11][12][13][14][15][16][17][18][19][20][21][22][23][24][25][26][27], the references related to assessing cybersecurity [28][29][30][31][32][33][34][35][36][37][38][39], and references related to assessing both availability and cybersecurity as attributes of dependability of cloud and IoT systems [40][41][42][43].…”

“…The approach allowed estimating the probabilities of cloud threats and types of attacks, which were confirmed by the simulation results. To explore the relationship between physical servers' failure rates and job failure events, a CTMC reliability model for Google cluster physical machines was presented in [30]. The reliability model was utilized for evaluating steady-state availability, steady-state unavailability, mean time to failure, and mean time to repair in the Google cluster.…”