Selected Papers from the International Conference on Information, Communication, and Engineering 2013View this Special Issue
Research Article | Open Access
Bao Rong Chang, Hsiu-Fen Tsai, Chi-Ming Chen, "Empirical Analysis of Server Consolidation and Desktop Virtualization in Cloud Computing", Mathematical Problems in Engineering, vol. 2013, Article ID 947234, 11 pages, 2013. https://doi.org/10.1155/2013/947234
Empirical Analysis of Server Consolidation and Desktop Virtualization in Cloud Computing
Physical server transited to virtual server infrastructure (VSI) and desktop device to virtual desktop infrastructure (VDI) have the crucial problems of server consolidation, virtualization performance, virtual machine density, total cost of ownership (TCO), and return on investments (ROI). Besides, how to appropriately choose hypervisor for the desired server/desktop virtualization is really challenging, because a trade-off between virtualization performance and cost is a hard decision to make in the cloud. This paper introduces five hypervisors to establish the virtual environment and then gives a careful assessment based on C/P ratio that is derived from composite index, consolidation ratio, virtual machine density, TCO, and ROI. As a result, even though ESX server obtains the highest ROI and lowest TCO in server virtualization and Hyper-V R2 gains the best performance of virtual machine management; both of them however cost too much. Instead the best choice is Proxmox Virtual Environment (Proxmox VE) because it not only saves the initial investment a lot to own a virtual server/desktop infrastructure, but also obtains the lowest C/P ratio.
Well-known public large-size cloud and private enterprise-owned cloud are currently leading cloud computing and services, for example, Amazon AWS, Google App Engine, and Windows Azure in public cloud and Saleforce.com and EMC2 in private cloud. In contrast, small-to-medium sized enterprises (SMEs), educational institutes, and social groups are also very eager to pursue the services they want based on virtual server in cloud (VSiC)  architecture due to cost reduction, performance escalation, and security improvement. With this service, the operational costs for the information system can be drastically reduced and it can quickly increase the competitiveness of its information system, which is sustained by the following advantages: centralized monitoring, quick management, dynamic optimization, and efficient backup.
Technically, unexpected situations with service-type servers, such as websites, databases, AP servers, and file servers, bring much trouble for enterprises. Once a service stops its function, it can cause faulty data, stalled production lines, and interrupted operation procedures, leading to multiple losses. However, a physical host has problems of promptness of service transfer to another host, restarting the service, and inability to update data in real time. The hardware, information, and data will be obstacles for enterprises to overcome.
To solve the issues mentioned above, virtual machine management system or hypervisor , such as VMware ESX/ESXi Server, Microsoft Hyper-V R2 , and Proxmox virtual environment (Proxmox VE) , is able to deliver the virtual machine services for virtual server/desktop infrastructure with high availability in computing, secured networking, and consolidated storage. With this approach, users make it possible to adopt low cost thin clients (a low-end PC or PDA) to link to the system for the services , reducing IT purchasing cost and saving computation power because thin client devices are easier to setup where the chances of malfunction, heat crash, and computer virus are considerably low. In addition, easy to use through wireless mobile computing environment gains peoples’ attraction a lot.
As the virtualization architecture expands continuously, network storage services have become part of the virtualization architecture. Openfiler  is a storage management system used for disk read/write accessing in a shared storage among several virtual machines or servers in enterprises. It is a free and conducive system that supports both network-attached storage (NAS)  and storage area network (SAN)  functions. As installed, it can be managed via web browsers in conjunction with iSCSI shared storage (IPSAN)  technique to provide file accessing on cloud computing servers. One can also use logical unit number (LUN)  through iSCSI to complete the placement of block accessing for virtual machines that are created by VMware or Hyper-V. This paper will evaluate the performance of accessing to block storage area network with Openfiler.
The goal in this paper also wants to clarify a credibility of cost and benefits on infrastructure virtualization. Speaking of virtualization cost and benefits, we will give the exploration of total cost of ownership (TCO) and return on investment (ROI) individually in the following statements. ROI  related to two factors, savings and investment, is equal to savings/investment, where investment represents the sum of incremental investment in transition from physical to virtual (new servers, shared storage, software licenses & support, services and training, etc.). In addition to investment, TCO  yet includes IT administration and downtime cost. Apparently, ROI and TCO can be properly undertaken well according to how big infrastructure has been virtualized. In terms of virtualization, how many infrastructures can be virtualized that intuitively depend on the ratio of the virtual machines per core (VMs/core), the so-called consolidation ratio . The consolidation ratio is a measurement unit that virtualization vendors use with extreme prudence to provide a rough idea of the server consolidation level that can be achieved on their hypervisors. Virtual infrastructure as we know has been classified into server virtualization and desktop virtualization. We will explore the consolidation ratio and TCO/ROI for both server and desktop virtualizations in this study.
In addition, how to appropriately choose the virtual machine management (i.e., hypervisor) for the desired server/desktop virtualization is a really tough problem of a trade-off between performance and cost before making the decision to deploy virtual server in cloud (VSiC) as a new IT. Given five hypervisors used to establish VSiCs, this paper introduces a composite index to represent the evaluated functional performance as well as formulates a proper equation to stand for the estimated virtualization cost so that a C/P ration will conclude a careful assessment about a variety of VSiCs discussed in this paper.
2. Structure of Virtual Machine and Shared Storage
The purpose of this session is to setup five different virtual servers in cloud (VSiC) and provide an appropriate assessment for every virtual machine monitor (i.e., hypervisor). These assessments can provide an optimal solution for SMEs, schools, and social groups. This research will setup and implement five heterogeneous virtual machine management systems which are (a) vSphere ESX/ESXi, (b) Hyper-V R2, (c) Proxmox virtual environment, (d) Ubuntu KVM , and (e) CentOS Xen , where they are shown in Figures 1, 2, 3, 4, and 5, respectively. Moreover, five heterogeneous hypervisors all link to a shared storage Openfiler through LAN, which is a type of IPSAN storage as shown in Figure 6.
Noted that vSphere ESX/ESXi Server, Hyper-V Server 2008 R2, or CentOS-based Xen required at least a stand alone machine for installation, Ubuntu Enterprise Server has to at least include 2 physical machines (Cloud Controller and Node Controller) and Proxmox VE at least a master or optionally adding multiple nodes as well.
As shown in Table 1, hypervisor includes all kinds of virtual machine architectures and types, hence referred to as heterogeneous virtual server in cloud. Virtual machine architectures are divided into hosted architecture, such as (a) and (b), and bare-metal architecture, such as (c), (d), and (e); its types are classified into paravirtualization, such as (d) and (e), full-virtualization, such as (a), (b), and (c), and hardware-assisted virtualization for all of them. This is because the new x86 machines, regardless of 32-bit or 64-bit, now support Intel VT-x and AMD-V virtual instructions.
3. Formulation of Consolidation Ratio and TCO/ROI
The aim of this section is first to understand the consolidation ratio of VMware ESX server as well as TCO/ROI evaluated at VMware TCO/ROI calculator . Consolidation ratio means the number of VMs running in a server concurrently depending on the number of workloads and the average number of VMs per core. The max consolidation ratio per VMware ESX server is by default calculated as 1.5 VMs per core multiplied by the total number of cores per server . That is, it gives 12 : 1 in ESX’s server favor. Based on Taneja Group observations early in 2009  during testing as well as their familiarity with a broad range of virtual server infrastructures, they claimed there are many realistic workloads under which ESX4 gains a 2 : 1 VM density advantage comparing with Hyper-V R2 and XenServer 5.5. Thus, we extensively proposed to analyze VM density according to a VMware official document for EXS server  and a testing report from consulting services  such that the consolidation ratio for the other hypervisors can be obtained.
On the other hand, the ROI required for the transition to virtual infrastructure evaluates the percent of total saving/total cost, where total saving consists of capital expenditure, operational expenditure, and downtime cost; total cost is composed of new servers, storage, network storages, software license & support, server, and training . TCO is the costs associated with operation of datacenter which include capital expenditure (servers, storage, and switches), operational expenditure (power & cooling, infrastructure administration labors, and rack space), and business agility (planned downtime, unplanned downtime, and business downtime) . Furthermore, the TCO/ROI for the other hypervisors has been estimated carefully according to VM density and the ratio of ESX normalized performance index to anyone.
4. Empirical Analysis Method
With respect to the performance evaluation for the virtual machine monitor, a variety of guest OS and two well-known test tools are adopted in this study. PassMark PerformanceTest 7.0 (at http://www.passmark.com/download/pt_download.htm) is applied to the test of virtual machine performance for the Windows series guest OS like Windows XP, Windows 7, and Windows Server 2003, and UnixBench 5.1.3 (at http://byte-unixbench.googlecode.com/files/UnixBench5.1.3.tgz) is employed for Linux series guest OS like Ubuntu. According to evaluated performance score for each virtual infrastructure server, we derived the respective scores into a composite index, each hypervisor on (1) and (2), and sequentially normalized to be a value ranging from 0 to 1 on (3), where we refer to this as a normalized composite index related to virtual machine performance. In (1), is a test score for various guest Windows OSs (e.g., Win XP, Win Server 2003, and Win 7) running in a VM, and accordingly, represents a mean test score for various guest OSs; in (2), stands for a mean test score for a guest Linux OS (e.g., Ubuntu and CentOS); in (3), means a VM composite index for a specific hypervisor, and two coefficients and act as a weighted average; in (4), represents a normalized VM composite index for a specific hypervisor. Consider
In order to achieve virtual infrastructure together with a shared storage, we first have to establish a set of block storage area network system called Openfiler and then mount shared storage to each virtual server. After that, we go for the performance evaluation of accessing block storage by using Linux hdparm command  to test disk reading speed. Likewise, we do the same thing as the above mentioned procedure to develop a composite index on (5) and its normalized composite index on (6) associated with storage-accessing speed performance. Finally, we derive the composite index for total on (7) and a normalized one on (8). In (5), is a mean test score for cache read speed performance on a guest Linux OS in a VM, a mean test score for disk read speed performance, and a storage-accessing speed composite index; in (6), stands for a normalized storage-accessing speed composite index for a certain hypervisor; in (7), represents a composite index for a certain hypervisor for overall, and in (8), represents a normalized composite index. In terms of average, three sets of coefficients indicated by , , and are designated to act as a weighted average for equation in (5), (7), and (8), respectively. Consider
In this paper, we mainly conduct a credibility of cost and benefits before and after infrastructure virtualization. Speaking of cost and benefit, we will explore the consolidation ratio and TCO/ROI of both server consolidation and desktop virtualizations as mentioned above. VMware ESX server is first chosen to evaluate its consolidation ratio and estimate TCO/ROI at VMware calculator webpage. The consolidation ratio and TCO/ROI of both server and desktop virtualizations for the other hypervisors will be proportional to both its VM density (major part) and the ratio of ESX normalized composite index to alternative one (minor part). We broke the costs about capital expenditure, operational expenditure, and business agility into 13 items. Technically the table as listed in Table 2 gives us an insight to realize which item highly concerned with the VM density and/or normalized performance index ratio, which are derived from TCO/ROI calculation. Moreover, a formula for calculating the expenditure of TCO/ROI has been derived on (9) where represents VM density for ESX server and VMDhypervisor for the other hypervisors; CostESX stands for the expenditure for ESX server and Costhypervisor for the other hypervisors. There is no the initial cost of software package for Proxmox VE, Ubuntu KVM, and CentOS Xen due to open source software. However, the initial cost of ESX server software package (approximate US$ 12,668) is greater than that of Hyper-V R2 (approximate US$ 6,000). Consider
5. Experimental Results and Discussion
There are three experiments and the discussion presented in the following subsections.
5.1. Assessment of Virtual Machine Performance
The server hardware specification is listed in Table 3.
The resulting score is an average of various scores from test items, for example, CPU, memory, storage, network, and 2D graph. In the experiment, two testing softwares (PassMark PerformanceTest 7.0 and UnixBench 5.1.3) are applied to evaluate the virtual machine performance for hypervisors such as ESXi 5.0, Hyper-V R2, Proxmox VE, Ubuntu Enterprise Server KVM, and CentOS Xen. A summary of the virtual machine performances is shown in Table 4 as well as Figures 7 and 8.
Different guest OS is installed separately on each virtual machine and adds up to 5 VMs in each predetermined virtual machine monitor (one of the above-mentioned hypervisors). They are divided into Windows series and Linux series guest OSs. After that, the testing software will be taken to analyze a set of target items as mentioned above.
5.2. Performance Evaluation of VM Accessing Shared Storage
In the experiment, according to the same server hardware specification as mentioned above, Openfiler storage device is separately mounted onto five virtual servers in cloud to test disk-accessing speed. Tests carried out two disk-accessing speed indicators, (a) timing cached reads (MB/sec) and (b) timing buffered disk reads (MB/sec) . A summary of storage-accessing speed performances is shown in Table 5 as well as Figures 7 and 8.
5.3. Estimation of Consolidation Ratio and TCO/ROI
This part goes to the estimation of the consolidation ratio and TCO/ROI for each virtual server infrastructure as mentioned above. We choose VMware ESXi server as a benchmark and use VMware TCO/ROI calculator to yield its server/desktop consolidation ratio and TCO/ROI quantity. VMware ESX exceeds 2 times the workload capacity per server for the competitive hypervisors such as hypervisor-V R2, XenServer 5.5, and the others . Accordingly, we assumed ESXi achieved 2 : 1 VM density per server advantage over the other hypervisors in this study. Thus, ESXi achieved the consolidation ratio 12 : 1 in server virtualization as well as 96 : 1 in desktop virtualization because of 1.5 VM per core; instead the other hypervisors result in 6 : 1 in server virtualization as well as 48 : 1 in desktop virtualization. We estimate that TCO/ROI for alternative hypervisor is greatly proportional to its VM density (i.e., major part of estimation) and additionally adds somewhat fluctuations or changes to the evaluation of TCO part based on the ratio of ESXi normalized composite index to alternative one (i.e., minor part of estimation). In the experiment, we have deployed up to 50, 100, 150, 200, 250, and 300 workloads for the transition to their respective server virtualization with a 5-year duration. A summary of TCO/ROI calculation for 5 years is shown in Figures 9 and 10 in server virtualization as well as Figures 11 and 12 in desktop virtualization.
5.4. Cost Estimation for Server/Desktop Virtualization
According to the cost formula on (9), this part gives the operation cost for server/desktop virtualization so that the ratio of cost to performance (C/P ratio) for various virtual machine managements (hypervisors) is able to carry out the goal of the assessment in this paper. With a basis cost of vSphere ESX server indicated in (9), the operation cost computation to Hyper-V R2, Proxmox virtual environment, Ubuntu KVM, and CentOS Xen are conducted to disclose the real assessment of virtual server/desktop infrastructure. In terms of VM performance as indicted in (7) and (8), the composite index presents how the function of VM and its related storage-accessing speed perform in the various hypervisors, and the results are shown in Figures 7 and 8 as well. As we can see in Figure 13, a Costhypervisor divides and yields the C/P ratio with varying coefficients α/β, and the summary is also listed in Table 6. Even though Hyper-V R2 gains the best performance, Proxmox VE obtains the best choice in this study due to the lowest C/P ratio.
In terms of VM performance and its storage-accessing speed as shown in Figures 7 and 8, Hyper-V R2 gains the best performance overall because of the highest composite index. As shown in Figures 9 and 10 in the server virtualization, the best choice is to take 150 workloads with ESX server because this combination will achieve the highest ROI and the lowest TCO over the duration of 5 years. However, when we look at desktop virtualization as shown in Figures 11 and 12, Proxmox VE outperforms the others to achieve the best ROI over a 5-year duration, even though its TCO is a little bit less than ESX’s TCO. In addition, Proxmox VE would not only be the best choice to save the initial investment to own a virtual server infrastructure with better ROI in desktop virtualization, but it also obtains the lowest C/P ratio. This paper does not mention the security issue for virtual machines, which is related to access control  and cryptograph in VMs . It can be explored in the further work.
The objective of this paper is to explore several critical issues of virtual server/desktop infrastructure such as server consolidation, virtualization performance, virtual machine density, total cost of ownership (TCO), and return on investments (ROI). Thus, this paper introduces five distinct well-known hypervisors installed in VSiC and has proceeded with an empirical analysis of server consolidation and desktop virtualization. As a result, even though ESX server gets the highest ROI and the lowest TCO in server virtualization and Hyper-V R2 gains the best performance over all, both of them cost too much; instead Proxmox VE would not only be the best choice to save the initial investment to own a virtual server infrastructure with better ROI in desktop virtualization, but also it obtains the lowest C/P ratio. We drew the conclusion that Proxmox VE outperforms the other hypervisors operating in the virtual server/desktop infrastructure.
This work is supported by the National Science Council, Taiwan, under Grant no. NSC 100-2221-E-390-011-MY3.
- B. R. Chang, H. F. Tsai, and C. M. Chen, “Evaluation of virtual machine performance and virtualized consolidation ratio in cloud computing system,” Journal of Information Hiding and Multimedia Signal Processing, vol. 4, no. 3, pp. 192–200, 2013.
- J. Hardy, L. Liu, C. Lei, and J. X. Li, “Internet-based virtual computing infrastructure for cloud computing,” in Principles, Methodologies, and Service-Oriented Approaches For Cloud Computing, chapter 16, pp. 371–389, IGI Global, Hershey, Pa, USA, 2013.
- Y. Haga, K. Imaeda, and M. Jibu, “Windows server 2008 R2 hyper-V server virtualization,” Fujitsu Scientific and Technical Journal, vol. 47, no. 3, pp. 349–355, 2011.
- A. Kovari and P. Dukan, “KVM & OpenVZ virtualization based IaaS open source cloud virtualization platforms: openNode, proxmox VE,” in in Proceedings of IEEE 10th Jubilee International Symposium on Intelligent Systems and Informatics, pp. 335–339, 2012.
- S. M. S. Kimball and C. Bisciglia, “Cluster computing for web-scale data processing,” ACM SIGCSE Bulletin, vol. 40, no. 1, pp. 116–120, 2008.
- F. Guthrie, S. Lowe, and M. Saidel-Keesing, VMware vSphere Design, John Wiley & Sons, New York, NY, USA, 2011.
- D. Sacks, “Demystifying DAS, SAN, NAS, NAS Gateways, Fibre Channel, and iSCSI,” IBM Storage Networking, pp. 3–7, 2001.
- I. Yakti and W. A. Salameh, “iSCSI (internet small computer system interface) for your startup,” International Journal of Information and Communication Technology Research, vol. 20, no. 3, 2012.
- J. M. Kurz, “Beyond ROI,” Value Management, pp. 1–7, 2012.
- M. Kornevs, V. Minkevica, and M. Holm, “Cloud computing evaluation based on financial metrics,” Information Technology and Management Science, vol. 15, no. 1, pp. 87–92, 2013.
- A. Beloglazov and R. Buyya, “Optimal online deterministic algorithms and adaptive heuristics for energy and performance efficient dynamic consolidation of virtual machines in Cloud data centers,” Concurrency and Computation, vol. 24, no. 13, pp. 1397–1420, 2012.
- S. Simanta, G. A. Lewis, E. J. Morris, K. Ha, and M. Satyanarayanan, “Cloud computing at the tactical edge,” Software Engineering Institute 680, 2012.
- B. K. Kim, K. W. Hur, J. H. Jang, and Y. W. Ko, “Feedback scheduling for realtime task on Xen virtual machine,” Communications in Computer and Information Science, vol. 266, no. 2, pp. 283–291, 2011.
- VMware TCO/ROI Calculator, VMware, http://roitco.vmware.com/vmw/.
- Frequently Asked Questions, Report of VMware ROI TCO Calculator, Version 2.0, VMware, 2013.
- TANEJA Group Technology Analysis, “Evaluating the ESX 4 Hypervisor and VM Density Advantage,” Technology Validation, pp. 1–12, 2009.
- M. Tabuchi, K.-I. Itoh, Y. Nomura, and H. Taniguchi, “Design and evaluation of a system for running two coexisting linux systems,” Electronics and Communications in Japan, vol. 90, no. 9, pp. 56–68, 2007.
- M. Aboutabl, A. Agrawala, and J. D. Decotignie, “Temporally determinate disk access (extended abstract): an experimental approach,” ACM SIGMETRICS Performance Evaluation Review, vol. 26, no. 1, pp. 280–281, 1998.
- S. Prabhakar, S. Pankanti, and A. K. Jain, “Biometric recognition: security and privacy concerns,” IEEE Security and Privacy, vol. 1, no. 2, pp. 33–42, 2003.
- R. L. Krutz and R. D. Scott Lowe Vines, Cloud Security: A Comprehensive Guide to Secure Cloud Computing, John Wiley & Sons, New York, NY, USA, 2010.
Copyright © 2013 Bao Rong Chang et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.