CLOUD computing

Abstract

CLOUD computing (Grid or utility computing, computing on-demand) which was the talk of the computing circles at the end of 1990s has become once again a relevant computational topic. CLOUD computing, also considered as a fifth utility after water, electric power, gas, and telephony, is on the basis of the hosting of services on clusters of computers housed in server farms. This article reviews CLOUD computing fundamentals in general, its operational modeling and quantitative (statistical) risk assessment of its much neglected service quality issues. As an example of a CLOUD, a set of distributed parallel computers is considered to be working independently or dependently, but additively to serve the cumulative needs of a large number of customers requiring service. Quantitative methods of statistical inference on the quality of service (QoS) or conversely, loss of service (LoS), as commonly used customer satisfaction metrics of system reliability and security performance are reviewed. The goal of those methods is to optimize what must be planned about how to improve the quality of a CLOUD operation and what countermeasures to take. Also, a discrete event simulation is reviewed to estimate the risk indices in a large CLOUD computing environment favorably compared to the intractable and lengthy theoretical Markov solutions.

Assessing the service quality in Software-as-a-Service from the customers’ perspective: a methodological approach and case of use

“…Google, IBM, Microsoft, HP, Amazon, and Yahoo) have already made investments not only in cloud research but also in establishing cloud computing infrastructure services (Sahinoglu & Cueva-Parra, 2011). By 2018, more than 78 percent of workloads will be processed by cloud data centres (22 percent will be processed by traditional data centres), and, 59 percent of the total cloud workloads will be SaaS workloads, up from 41 percent in 2013 (Cisco, 2014).…”

Modeling and simulation in engineering

“…This fact is not only because of involvement of human life and death situations in adversity, as compared to accidental casualties in the production world, but also because of lack of theoretical and experiential data base dating back to only 1990s since the launch of public internet. The author, by following examples in this area proceeds with currently popular VaR technique by Kim et al49 and Security Meter and CLOUD simulation tools (CLOURA) by Sahinoglu et al26 Monte Carlo VaR is costly to execute; it does not incorporate cost comparisons when taking measures. Consider a medium size firm holding positions in 20,000 different financial instruments.…”

“…The capital cost of hardware plus the operational cost for data center space, power, cooling and maintenance makes this cost prohibitive to all but the largest firms. However, scalable CLOURA is a very fast algorithm, that is, it can simulate a CLOUD system with 430 servers for 1000 years in less than 4 min 26 . SM simulations as in Tables 7–9 and Figures 15–17 and are relatively fast and accurate, comparable to their analytical counterparts 20–22…”

“…Figure 1 displays the interaction between the process of building a model by focusing on the interplay between (1) experimental results, (2) simulation results, and (3) theoretical predictions as displayed in Ref 25. A favorable example of this interplay is presented in a recent WIREs article titled Cloud Computing 26 (Figure 8, p. 55), which displays an experimental scenario for a trivial Cyber Cloud. On the other hand Ref 26 (Figure 9, p. 57) outlines the Markovian theoretical predictions followed by the simulation results for the same scenario of two 1‐GB units serving a constant load of 1.5 GB for 13 cycles.…”

“…A favorable example of this interplay is presented in a recent WIREs article titled Cloud Computing 26 (Figure 8, p. 55), which displays an experimental scenario for a trivial Cyber Cloud. On the other hand Ref 26 (Figure 9, p. 57) outlines the Markovian theoretical predictions followed by the simulation results for the same scenario of two 1‐GB units serving a constant load of 1.5 GB for 13 cycles. The resulting availability of this small Cloud: (1) 0.307 for Experimental , (2) 0.305 for Simulation after 1 million runs, or trials and (3) 0.331 for Markov Theoretical , allowing a negligible error content, which diminishes to less than 3% as the size of the experiment increases from a few to many hundreds of units.…”