Performance Modelling and Analysis of an OpenStack IaaS Cloud Computing Platform

Abstract

Performance is one of the main aspects that should be taken into consideration during the design, development, tuning and optimisation of computer networks supported by cloud computing platforms (CCPs). Queueing network models (QNMs) of CCPs constitute essential quantitative tools of investigation towards identifying acceptable levels of quality-of-service (QoS), whether for upgrading an existing CCP or designing a new one. In this paper, a new stable open QNM with either single or multiple server queueing stations, first-come-first-served (FCFS) scheduling and random routing is proposed for the performance modelling and analysis of an OpenStack Infrastructure as a Service (IaaS) CCP. In this context, it is assumed that the external arrival process is Poisson and the queueing stations provide exponentially distributed service times. Based on Jackson's Theorem, the open QNM is decomposed into individual M/M/c queues with c server(s) (c ≥ 1) and exponential inter-arrival and service times, each of which can be analysed in isolation. Consequently, closed form expressions for key performance metrics of the QNM are determined, such as those for the mean response time, throughput, server (resource) utilisation and the probability of the number of requests by clients at each queueing station during waiting for and/or receiving resource provisioning. The evaluation of these metrics identifies the bottlenecks of the CCP that are causing the worst network delays and associated performance degradation and thus, provides insights into the capacity planning of networks with OpenStack IaaS solutions for CSPs.