Court Filings

Cost and Performance Analysis of Network Function Virtualization based Cloud System

As the users on the cloud network increase, the consumption of the Compute, Network and Storage resources also increases. This leads to increase in the cost of deployment, configuration and maintenance. Hence, the capital expenditure (CAPEX) of the
of 5
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  Cost and Performance Analysis of Network Function Virtualization based Cloud Systems Ananth M. D. and Rinki Sharma Department of Computer Engineering, M.S.Ramaiah University of Applied Science, Bangalore, India and  Abstract - As the users on the cloud network increase, the consumption of the Compute, Network and Storage resources also increases. This leads to increase in the cost of deployment, configuration and maintenance. Hence, the capital expenditure (CAPEX) of the organization providing the cloud network increases. Network function virtualization (NFV) is a technology which virtualizes network functionalities. This paper studies the influence of NFV on CAPEX of cloud based networks and compares it with traditional implementation (without NFV) of such networks. A prototype cloud network based on NFV implementation is developed and implemented. Based on the test cases developed on the prototype, CAPEX of the resources used for both NFV based and traditional implementations are studied and analyzed. RESTful web services are created for the users of the cloud network to orchestrate and manage the network on the cloud. The results accomplished show that NFV based implementation reduces the CAPEX, when compared with the traditional implementation. It is also observed that orchestration mechanism reduces complexity of management of cloud network. This paper also studies the performance of the cloud and physical system in terms of load on the CPU. It is observed from the analysis that the behavior and performance of these systems is similar.  Keywords- Cloud; Cloud Management; Network Function Virtualization; NFV; CAPEX; OpenStack; Orchestration I.   I NTRODUCTION   Internet has become an integral part of everyone’s life in present days. The services are delivered by the service providers, as requested by the end user. The evolution of cloud computing has provided an opportunity for the end user to customize the use of the internet as required and pay only for the requested services. However, to provide all these simplicities and customization, there are a few complex systems that work behind the scenes. Every vendor providing services to the customers has a data centre to host the services and manage them. There are severs, routers, switches and middleboxes running in a data centre environment. In traditional networks, there are physical dedicated hardware resources for compute, network and storage components, separately performing their respective tasks [1]. Increased complexity of these networks leads to increase in cost. Administration and management of the physical devices in these networks increases the capital expenditure (CAPEX) and operational expenditure (OPEX) for a service provider. More number of physical devices, higher the CAPEX. Moreover, network resource usage is highly unpredictable, leading to uncertainty among the service providers [1]. Not all network functions or the internetwork devices are used round the clock. Some of the network functions and devices are unused or not fully utilized. These devices and functions can be managed appropriately by sharing the resources with other users or suspending it when not in use. The solution for the traditional network management is virtualization of compute, network and storage resources [2]. NFV virtualizes the resources such as compute, network and storage. It allows the implementation of these resources as and when needed on a hardware infrastructure. The software implementations of the network function can be decoupled from the corresponding physical resources and infrastructure through NFV [2]. This enables the functionality of the network functions such as routing, switching, firewall etc., to run on a generic hardware which can be virtualized and shared between multiple systems hence reducing CAPEX [1]. This paper presents the difference in CAPEX between traditional network and the cloud network using NFV. Virtualization of network functions using NFV is carried out on the cloud computing platform. Rest of this paper is organized as follows. Section II presentssurvey of work related to the NFV cloud management. In Section III implementation of the prototype and the process flow are explained. In Section IV, calculation of CAPEX based on the developed prototype and the test cases are discussed. In Section V, performance of the Cloud system is compared with the physical environment using a use case scenario. Section VI discusses the results obtained from the test cases. Section VII concludes the paper and discusses future work. II.   R ELATED W ORK  The advancement in the distributed systems and internet technology gave way for cloud computing, where the hardware resources were placed at a centralized data centre and the users could access the resources through internet [3]. Data centres comprise of high end resources for computing thus reducing the burden of resource maintenance at users’ end [3]. Cloud infrastructure provides framework for manageable, scalable and reliable applications. Most of the  cloud computing strategies implement virtualization, where the hardware resources namely the compute resources such as random access memory (RAM), processor, hard disk, network interface cards (NIC) etc., are shared between two or more clients. In traditional/legacy networks the network functions are tightly coupled with the infrastructure they run on. NFV decouples the software implementation of the network functions from the resources (such as computation, storage and networking) they use, thus providing flexibility to move network functions from dedicated devices to generic servers. Thus NFV is considered to be economical, scalable, flexible and secure technology for shaping present day networks [2]. A framework for management and orchestration (MANO) of resources in the cloud data centre is defined by the European Telecommunication Standards Institute (ETSI). The ETSI NFV-MANO document discusses the management and orchestration framework for the provisioning of virtualized network functions (VNF) [2]. OpenStack is a cloud operating system that controls compute, network and storage components [4]. Relationship of NFV with software defined network (SDN) and cloud is discussed in [1]. NFV abstracts network functions, SDN abstracts the network infrastructure and the cloud computing abstracts the compute infrastructure (such as CPU, RAM, etc.). It will be beneficial to use these three technologies since they share similar advantages of virtualization, cost reduction, agility, automation and dynamism [1]. The benefits of using OpenStack as the API framework for NFV are discussed in [5]. Several use cases of NFV are discussed in [6]. The authors in [6] consider reliability, stability and security asthe key performance parameters in both physical and in software based virtualized networks. The concept of network function centre (NFC), a cloud platform that delivers network services on subscription basis to the clients is discussed in [7]. The paper identifies that dynamic scaling up and down of resources for network resources upon the request from clients is a challenge. A Network Function enabled multi-tenant Cloud architecture, called NeFuCloud is presented in [8]. NFV is used to virtualize the middleboxes in the multi-tenant cloud. The architecture follows the principle of SDN to separate the control and data plane, thus providing flexibility, interoperability and adaptation to various strategies and policy changes. The authors in [8] consider network function deployment and policy enforcement, performance guarantee and isolation, and resource management as the major challenges for multi-tenant clouds. To study theeffect of network virtualization on CAPEX in a data centre, the authors in [ 9 ] have considered two deployment strategies namely vertical serial deployment (VSD) and horizontal serial deployment (HSD). A framework for virtual network function (VNF) management and orchestration in enterprise wireless local area network (WLAN) is presented in [10]. The goal of the work presented in [10] is to reduce latency in the wireless network through VNF. The work presented in this paper analyzes the CAPEX and OPEX on the cloud network prototype. The prototype is developed using NFV and OpenStack, on a physical infrastructure upon which the compute and network functions are virtualized. III.   D ESIGN AND D EVELOPMENT  Design of the cloud prototype is based on the ETSI NFV MANO Architectural framework [2]. The design of the prototype used in this paper is shown in Figure 1. The physical infrastructure consists of Intel core i7 5 th  Generation processor with 8 GB of RAM used as the NFV infrastructure (NFVI) on which the prototype is developed. OpenStack [4] is used as virtual infrastructure manager (VIM), to create and manage network and compute resources. Virtual routers and machines provided by OpenStack are used to create VNF in the test cases. An orchestrator is developed using Python based pyramid web framework [11] and SQL Alchemy [12], which exposes the RESTful services [11]. Through the RESTful services, users can create, deploy, and delete, the networks and network functions. Figure 1: Design of Cloud Management using NFV User inputs the requirements and configures the required cloud components through REST API’s.  These inputs are then stored on to the database. Pyramid web framework is used with Alchemy scaffold to take the inputs from the user through REST API’s and store it to the database.  The dat abase is designed to have three tables, namely ‘vld’, ‘vnfd’ and ‘nsd’, where vld, vnfd and nsd stands for virtual link descriptors, virtual network function descriptors and network service descriptors, respectively. The ‘vld’  table consists of the details of the network name, subnet IP addresses and gateway IP address. The ‘vnfd’ t able comprises of the details of the name of VNF, name of the operating system image or VNF image file and the type of the hardware it is using to run the image. The type of the hardware is mentioned as Flavours. For example, flavour by the name ‘m1.small’ consists of one virtual CPU (vCPU), 2 GB of RAM and 20 GB of hard disk. The ‘nsd’  table consists of the relationship between the ‘vld’  and ‘vnfd’  components. In other words, nsd consists of the topology of the network with the interconnections between vld and vnfd. The values in the database are used to create a file in YAML (YAML Ain’t Mark  -up Language) format. A python script is run to create this file. The python script reads the data from  database, converts it into YAML format, and generates a file.The file contains the information of vld, vnfd and nsd that is present in the database.This file is provided as the input to the orchestration module of the OpenStack. IV.   T EST C ASES AND C ALCULATION OF CAPEX    A.   Test Cases: Three test cases are developed to analyze the CAPEX. First test case has two private networks and one VNF instance, as shown in Figure 2. Second test case has three private networks and two VNF instances, as shown in Figure 3. The third test case has two private networks, with two VNFs each, and the two networks are connected with a router, as shown in Figure 4. All the three network topologies of the test cases deployed on cloud running on single hardware platform using NFV is shown in Figure 5. Figure 2: Test Case 1 Figure 3: Test Case 2 Figure 4: Test Case 3 Figure 5: All the Test Cases  B.   CAPEX Calculation CAPEX is calculated based on the initial investment made for the devices that are used in the test cases. Similarly, OPEX is calculated based on the cost incurred to maintain and operate the devices on a day to day basis. The cost of the essential devices such as RAM, processor, motherboard, hard disk drive (HDD), L2/L3 Network Switches are obtained for CAPEX from the resources in [13, 14, 15]. To compute CAPEX, popular brands of the hardware, such as Intel, Cisco, D-Link, Kingston, Transcend and Seagate are considered. The average costs of the devices for physical implementation in a traditional network for all the three test cases are tabulated in Table 1, while, the costs of the devices on which the NFV is implemented are tabulated in Table 2. The CAPEX of the devices for both the implementations are tabulated in Table 3. T ABLE 1:   C OST OF R ESOURCES USING T RADITIONAL I MPLEMENTATION   Resource Cost (in INR)  RAM 980 Processor 8500 Motherboard 6300 HDD 940 Cables 320 Switches 690 L2/L3 Switches 11500 T ABLE 2:   C OST OF R ESOURCES FOR NFV   I MPLEMENTATION   Resource   Cost (in INR)  RAM 2600 Processor 22650 Motherboard 5600 HDD 940 T ABLE 3:   CAPEX  FOR T EST C ASES   CAPEX (in INR) Physical Implementation NFV Based Implementation Compute Resource Virtualization Test Case 1 18740 31790 33810 Test Case 2 36790 31790 33810 Test Case 3 79760 31790 43930  V. P ERFORMANCE A NALYSIS USING A S IMPLE W EB S ERVER  A use case is developed to analyze the performance of the system on cloud as compared to the physical system. Linux based system running Ubuntu operating system is used as a Web Server, serving a simple web page with minimal contents as shown if Figure 6. The web server is run on two environments, one on cloud and the other as a physical system. On both the environments, the server is configured to run on 1 GB of RAM and with a single core CPU. The topology of the web server on cloud environment is as shown in Figure 7. Web Server is connected to a private network called as Server Network which has series IP address. This is in turn connected to the external network through the network called as Public. Public network is connected to the Server Network using a router named as R1. This topology enables access to the Web Server on private network on cloud from external LAN networks. Figure 6: Web Page Figure 6: Topology of Web Server on Cloud Users who access the web server are simulated using a tool called as Webserver Stress Tool [16]. This tool is run on a separate physical system that is connected to the system that is running web server through a cable on LAN. Number of users are increased gradually and the performance of the web server, based on the load on the CPU, is analyzed on both the environments. The simulation simulates users accessing the web server for a period of 60 seconds, and the average load on CPU for a period of 60 seconds in terms of percentage of usage is shown in Table. Tool called as Glances [17] is used to calculate the CPU load for the process used by the web server. Table 4: Load on CPU No. of Users Load on CPU (%) System on Cloud Physical System 4 0.3 0.31 40 0.3 0.31 400 0.5 1.04 2000 1.11 1.43 4000 0.65 0.97 VI. R ESULTS AND D ISCUSSIONS  Graphs are plotted for the values obtained in Table 3 for analysis of CAPEX. From the graph shown in Figure 6, it is evident that the cost of the CAPEX increases as the devices in the test cases increases. Test case 3 comprises of network devices such as L3 Switches for inter-domain routing, leading to increased CAPEX due to cost of network devices. In case of NFV, all the three test cases were implemented and run on the same hardware by virtualizing the compute and network functions. Hence, the graph shows a constant CAPEX valuefor all the three test cases. Figure 7: CAPEX Analysis Graph Apart from the traditional physical implementation and the NFV based implementation, there’s one more way of implementing the system which virtualizes only the compute resources, such as RAM, HDD etc., of the system, but not the network devices or functions (for example a VM Virtual Box environment). This is known as compute resource virtualization. The CAPEX value of the compute resource virtualization is steady in test cases 1 and 2, until a L2/L3 network device is introduced to the network topology in test  case 3. The CAPEX value of compute resource virtualization is lesser than the traditional implementation with physical devices, and greater than the NFV based implementation. Graph of performance of the system in terms of average CPU load on cloud and physical environments, tabulated in Table 4, is as shown in Figure 8. In both the environments, load on the CPU increases as the number of users increase. The graph shows that the performance and behavior of the web server is similar on both the environments. In this particular use case, with the defined hardware configuration, average load on CPU on cloud environment is lesser than that of its counterpart. Figure 8: Performance of CPU It is evident from the CAPEX analysis graph in Figure 7 that the cost of physical resources increases as the number of physical system increases on a network. Also, from the graph of performance of CPU in Figure 8, the performance and behavior of the web server on cloud and physical systems are similar. Therefore, cost to performance ratio of a network can be improved by replacing the existing physical devices, wherever necessary, with Cloud and NFV based systems. VII. C ONCLUSION AND F UTURE W ORK  The work presented in this paper implements a prototype of the NFV based cloud network system. An NFV based cloud network system virtualizes the network functions and other internetwork system devices such as servers, firewalls etc., as per the user requirement. From obtained results it is observed that NFV based system thereby reduces the CAPEX of an organization, while providing flexibility and ease of management. The performance of the system was analyzed by developing a use case and increasing the load on the web server. Users accessing the web server were simulated and the numbers of users were gradually increased to increase the load on the server. From the analysis of the performance of the web server, it was observed that the performance and behaviour of the web server on two environments was similar. The work presented in this paper can be extended to service function chaining (SFC) in future. Load on CPU can be refined further and analyzed by introducing Load Balancing mechanism and using dynamic web pages. Analysis of the performance of RAM can also be incorporated. Another field of future work is to enhance the Cloud security. As the number of functions increases on the cloud, the cloud provider should have a strong protection for the cloud. NFV can also be tried on the scientific computing platform. A scientific functionality running on a dedicated hardware can be virtualized using NFV. R EFERENCES   [1]   Mijumbi, R., Serrat, J., Gorricho, J. L., Bouten, N., De Turck, F., & Boutaba, R. (2015). Network function virtualization: State-of-the-art and research challenges.  IEEE Communications Surveys & Tutorials , 18  (1), 236-262. [2]     ETSI, N. (2014). GS NFV-MAN 001 V1. 1.1 Network Function Virtualisation (NFV); Management and Orchestration. [3]   Hwang, K., Dongarra, J., & Fox, G. C. (2013).  Distributed and cloud computing: from parallel processing to the internet   of things . Morgan Kaufmann. [4]  (2016). OpenStack Docs: Current  . Retrieved April, 17, 2016, from [5]     Kavanagh, A. (2015). OpenStack as the API framework for   NFV: the benefits, and the extensions needed.  Ericsson  Review , 2 . [6]   Han, B., Gopalakrishnan, V., Ji, L., & Lee, S. (2015). Network function virtualization: Challenges and opportunities for innovations.  IEEE Communications Magazine , 53 (2), 90-97. [7]   Rankothge, W., Ma, J., Le, F., Russo, A., & Lobo, J. (2015, May). Towards making network function virtualization a cloud computing service. In 2015 IFIP/IEEE International Symposium on Integrated Network Management (IM) (pp. 89-   97). IEEE. [8]   Yu, R., Xue, G., Kilari, V. T., & Zhang, X. (2015). Network   function virtualization in the multi-tenant cloud.  IEEE  Network  , 29 (3), 42-47. [9]   Yousaf, F. Z., Loureiro, P., Zdarsky, F., Taleb, T., & Liebsch, M. (2015). Cost analysis of initial deployment strategies for   virtualized mobile core network functions.  IEEE Communications Magazine , 53 (12), 60-66. [10]     Riggio, R., Rasheed, T., & Narayanan, R. (2015, May). Virtual network functions orchestration in enterprise wlans.   In 2015 IFIP/IEEE International Symposium on Integrated  Network Management (IM)  (pp. 1220-1225). IEEE. [11]   The Pyramid Web Framework  —   The Pyramid Web Framework v1.7  . (2016). . Retrieved 5 May 2016, from [12]   The Pyramid Web Framework  —   The Pyramid Web Framework v1.7  . (2016). . Retrieved 5 May 2016, from [13] . (2016). Online Shopping: Shop Online for  Mobiles, Books, Watches, Shoes and More - . Retrieved 24 June 2016, from [14] (2016). - Electronics, Cars, Fashion, Collectibles, Coupons and More Online Shopping | eBay. Retrieved 24 June 2016, from [15]   Flipkart  . (2016). Flipkart  . Retrieved 24 June 2016, from [16]   Webserver Stress Tool - Performance, stress & load test. . (2016). . Retrieved 15 August 2016, from  [17]   Nicolas Hennion, a. (2016). Glances - An Eye on your system . . Retrieved 15 August 2016, from
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!