Modelling exogenous variability in cloud deployments

Authors:
Giuliano Casale;Mirco Tribastone
Affiliations:
Imperial College London, London, United Kingdom;Ludwig-Maximilians-Universität, Munich, Germany
Venue:
ACM SIGMETRICS Performance Evaluation Review
Year:
2013

Citing 8
Cited 0

Scaling for E Business: Technologies, Models, Performance, and Capacity Planning

Scaling for E Business: Technologies, Models, Performance, and Capacity Planning
TPC-W E-Commerce Benchmark Evaluation

Computer
Queueing Networks and Markov Chains

Queueing Networks and Markov Chains
A MAP fitting approach with independent approximation of the inter-arrival time distribution and the lag correlation

QEST '05 Proceedings of the Second International Conference on the Quantitative Evaluation of Systems
Runtime measurements in the cloud: observing, analyzing, and reducing variance

Proceedings of the VLDB Endowment
Fluid Analysis of Queueing in Two-Stage Random Environments

QEST '11 Proceedings of the 2011 Eighth International Conference on Quantitative Evaluation of SysTems
A Performance Study on the VM Startup Time in the Cloud

CLOUD '12 Proceedings of the 2012 IEEE Fifth International Conference on Cloud Computing
OFBench: An Enterprise Application Benchmark for Cloud Resource Management Studies

SYNASC '12 Proceedings of the 2012 14th International Symposium on Symbolic and Numeric Algorithms for Scientific Computing

Quantified Score

Hi-index	0.00

Visualization

Abstract

Describing exogenous variability in the resources used by a cloud application leads to stochastic performance models that are difficult to solve. In this paper, we describe the blending algorithm, a novel approximation for queueing network models immersed in a random environment. Random environments are Markov chain-based descriptions of timevarying operational conditions that evolve independently of the system state, therefore they are natural descriptors for exogenous variability in a cloud deployment. The algorithm adopts the principle of solving a separate transient-analysis subproblem for each state of the random environment. Each subproblem is then approximated by a system of ordinary differential equations formulated according to a fluid limit theorem, making the approach scalable and computationally inexpensive. A validation study on several hundred models shows that blending can save up to two orders of magnitude of computational time compared to simulation, enabling efficient exploration of a decision space, which is useful in particular at design-time.