Abstract
Cloud data centers (CDC) are an integral part of today's internet services. Enterprises and Businesses around the world rely heavily on data centers for their daily computation and IT operations. In fact, every time we search for an information on the internet, or we use an application on our smartphones, we access data centers. In CDC, most compute resources are represented as virtual machines (VMs) which are mapped into physical machines (PMs). Performance is often is a key metric for CDC. This paper presents a stochastic model based on queuing theory to aid in studying and analyzing performance in CDC. CDC platforms are modeled with an open queuing system that can be used to estimate the expected Quality of Service (QoS) guarantees the cloud can offer. We give numerical examples to show how the model estimates the number of required VM instances needed to satisfy a given the QoS parameters. In particular, we plot the response time, drop rate and CPU utilization while varying the incoming request arrival rate, and for different number of VM instances. We cross-validate our analytical model using a DES (Discrete Event Simulator). Our analysis and simulation results show that the proposed model is able to estimate the number of VMs needed to achieve QoS targets when varying the arrival request rate.
I. INTRODUCTION
A cloud computing infrastructures consist of services that are offered and delivered through a data center, that can be accessed from a web browser anywhere in the world [1]. Cloud computing providers offer computing resources (servers, storage, networks, development platforms, and applications) to users either elastically or dynamically, according to userdemand and form of payment [2].
V. CONCLUSION
In this paper, we presented an analytical model that can be used in studying the performance of CDC and is able to estimate accurately the needed number VMs to achieve a target QoS metric. We have considered the typical architecture in which a CDC houses a collection of PMs that will be used to run VMs and also LBs. Scenarios were presented to illustrate the usefulness of our analytical model-specifically, in determining the impact of the number of allocated VMs on key performance and QoS parameters which included response time, drop rate and CPU utilization. We cross-validated the results obtained from our analytical model with simulation results obtained from the popular JMT simulator. The simulation and the analysis results are in agreement and thus implying that, our analytical model is correct. As a future work, we plan to conduct experimental work of an elastic-scaling mechanism on a real-world CDC in which our analytical formulas derived in this paper are used to scale resources automatically to meet QoS targets in accordance to variable workloads.
