using Software Defined Networks ... specify the different link parameters like Bandwidth, Latency etc. ... Software Defined Network (SDN) is a new paradigm.
Network Management and Performance Monitoring using Software Defined Networks Veena S, Ram P. Rustagi, K.N.B.Murthy PES University, Bangalore, India (sveena, rprustagi, principal)@pes.edu Abstract - Nowadays Computer Networks are in constant evolution. New methodologies and techniques are required to manage the network. Software Defined Network (SDN) is a new approach to manage and maintain the network infrastructure. Mininet is the most popular emulator to understand and experience how SDN works. Our proposed research addresses the following problems. i) Enhance default version of Mininet to create the custom topologies in a user friendly manner without explicitly writing any programs. The users/researchers can even specify the different link parameters like Bandwidth, Latency etc. through a simple text file and create the network topology of their choice to test their innovative ideas and protocols. ii) Our next focus is to collect the information from the SDN controller, preserve it for historical analysis and present via a web interface. iii) Use SDN to reduce the Layer-2 broadcast traffic in Data Centres. This is being achieved by making use of hierarchical switch connectivity with pseudo MAC addresses. iv) Our last focus is the introduction of Cross Layer Optimization algorithms for better resource utilization in Data Centres. These new techniques help the network administrator and the end user to experience better application and network performance. Keywords – Address Resolution Protocol, Data Centre, Mininet, Openflow, Software Defined Network.
I. INTRODUCTION Software Defined Network (SDN) is a new paradigm where the network intelligent plane called the control plane is separated from the forwarding plane called the data plane. This separation allows one to implement the control plane in an external entity, called controller [6], and the data plane resides in the packet forwarding elements, called switches. In this architectural implementation, the controller gets the whole information about the network, which makes the network management easy. The communication between the controller and the networking elements (switches) is done by the openflow [1][4] protocol. As the intelligence is moved out of networking elements and gathered in the external controller, the cost of the switches comes down. As the whole information about the network is available in the central controller, the management and maintenance of the network becomes very easy. Use of this theme is the main motivation for our research work. We have categorised our research into four different subtopics which facilitates network management and performance improvement. A. For the implementation of the new ideas and innovations of the researchers, current operational network infrastructure cannot be used as it disturbs the operational
traffic. There is a need for a realistic test bed which resembles the real time network scenarios. A virtual network test bed like Mininet [8] fulfils this requirement. However, in its default version, implementing real life network topologies requires lot of programming effort. First focus of our research is to improve the Mininet tool without causing any performance impact to help the researchers in representing real time custom topologies. B. SDN controller maintains current information about the network in memory (data structure). It will be useful if this information (both current and historical) can be presented to the user in an easy to understand manner. That is the next focus of our research which helps in network management. C. The internet traffic in Data Centres is increasing everyday. Data centre prefers to use Layer-2 network to the extent possible for better switching performance. However, network sees a lot of Address Resolution Protocol (ARP) broadcast traffic to determine MAC address of the end host, as the Ethernet switches uses the broadcast mechanism for unknown unicast. The next focus of our research is to improve the network performance by reducing the ARP broadcast traffic in Data Centres. D. Currently while allocating Data Centre resources only the application needs are considered. However, by making use of networking characteristics while considering application needs one can improve the resource utilization. This will automatically improve the performance of the network infrastructure. This is the last focus of our research. II. RELATED WORK The openflow based network can be implemented in a single Laptop/Desktop using the Mininet Emulator. Mininet creates a virtual network which is scalable on a single laptop using Linux processes in network namespace. In [3] authors found satisfactory results for the comparison and analysis of four different projects using Mininet and Virtualization concept. In [5] Nikhil Handigol and et al. Uses a Container based emulation approach to replicate the results of 16 published networking papers using Mininet and found satisfactory results. These works suggest that the Mininet emulator is the best tool for understanding and implementing new SDN concepts. In the default version of Mininet extensive programming effort is required to create the custom topology and all the networks created will be having the default IP network address 10.0.0.0/8. In [10] we have shown
that the process of creating the network topology of user’s choice can be made simple by specifying the requirements in a simple text based configuration file. Use of configuration file eliminates the programming effort required for the creation of custom topology. It is also shown that the different IP network domains can be created by specifying the different network addresses in the same configuration file. In [9] we have extended our previous work by including the different link parameters like bandwidth, latency, packet loss etc. in the configuration file to replicate the real-life network topologies and shown that the performance of Mininet does not get affected even after these changes to the default version. A sample configuration file with link parameters is shown in Figure 1. Corresponding topology diagram is as shown in Figure 2.
Figure 1. Configuration for Custom Topology with Different Link Parameters
Figure 2. Custom Topology with Different link parameters
These realistic topologies can be used by the researchers to test their innovative ideas and protocols. In SDN, whatever communication happens in the underlying network is captured by the central controller. In order to analyse and improve the performance of networks, it is necessary to capture the aggregated information from the central controller. In [11] the SDN controller is enhanced by us to aggregate and store the network information. As hash tables are used to store this aggregated information, further
accessing of information becomes very easy. A web GUI has been developed to get the centralised view of the network data. This helps the network administrator to analyse the traffic flows in the network and plan for the allocation of the resources. III. PROPOSED RESEARCH Nowadays many cloud services are hosted in Data Centres. Users are continuously interacting with the data that resides in these Data Centres (servers). This increases the internet traffic within the Data Centre. Data centres prefer to use L2 switch for better performance which however is subject to high broadcast traffic. The proposed research is to introduce a mechanism to reduce the broadcast traffic within the Data Centre to improve the performance of networks by using the concept of Software Defined Networks. In [7] the authors discussed about a scalable, fault tolerant layer-2 routing and forwarding protocol PortLand, which explores the possibility of giving a single plug-and-play fabric to Data Centres. They proposed a new design with the introduction of hierarchical pseudo mac addresses (PMAC) to hosts, and a Fabric Manager to manage these addresses. They proved that this new design reduces a lot of broadcast traffic within the Data Centres. Modern Data Centres contains thousands of virtual machines (VMs) and millions of end hosts. To forward the packets within the Data Centres they use Ethernet and layer-2 switches. While forwarding the packets switches use a self learning mechanism to know on which link a host resides. The switches also broadcast the packets on all of its links for an unknown host. This generates a lot of broadcast traffic especially the ARP broadcast. In the proposed research we would like to reduce this layer-2 broadcast traffic by making use of PMAC concept implemented with SDN controller and Openflow. The SDN controller is enhanced to behave as a Fabric Manager which manages the hierarchical addressing scheme to reduce the subnet level ARP broadcast [2]. This helps in situations like VM Migration and also eliminates the need to run a spanning tree protocol within a layer-2 network. As an impact of virtualization technology, several applications are hosted on different VMs in Data Centres to satisfy millions of users around the world. Currently the Data Centres are not considering the network load while offering the services to the end users. In our proposed work we are exploring the possibilities of offering the services by considering the application needs as well as the network status and decide the best path for the user. To achieve this an intermediary abstraction layer is introduced between the Network layer and the Application layer as shown in Figure 3. The Application Control Gateway (ACG) collects the application requirements and the available Data Centre (DC) resources. The Network Control Gateway (NCG) collects the current network status. Both will exchange the information before deciding the best path for the user.
Figure 3. Three - Tier Architecture for Service Provisioning
The users can be categorised into different classes, understand their requirements and then try to allocate the Data Centre resources and network resources by keeping a watch on the available resources and the load on the network. This helps in improving the performance of the network significantly and thereby the application performance. IV. CONCLUSIONS AND FUTURE WORK The SDN controller and the Mininet emulator can be used to replicate the current day operational networks in a user friendly manner with the help of a configuration file. The aggregated information about the network can be gathered for the analysis of the network resources. Using the Pseudo MAC addresses one can reduce the ARP broadcast traffic in Data Centre networks. It is also possible to optimize the resource utilization by considering the application needs and the load on the network. We would like to continue our research in the area of SDN and Openflow. Currently we are working on flow tracking problem. SDN controller can be used to locate different flows in the network. By gathering the information about the response time, it can identify the paths that are heavily loaded. If a particular flow (path) is having more response time, the central controller can locate such paths and reallocate some of the flows to different paths which are lightly loaded and reduce the load on the network. This can be achieved by updating the forwarding tables of some of the switches by the controller which may help in improving the performance of the network. REFERENCES [1] Blueprint for Introducing Innovation into the Wireless Networks. Technical Report openflow-TR-1-4, 2009. [2] G. Vijay Teja, Karishma Sureka, Sai Gopal and Ram P. Rustagi. Application Development in SDN - Reduction of Broadcast Traffic in Data Centers. Project Report, 2014.
[3] Msahil M, Pujolle G, Serhrenchni A, Fadlallah A, Guenane F. “Openflow and on demand Networks.” Third international conference on “Network of the Future”. Nov. 2012. [4] Nick Mckeown, Tom Anderson, Hari Balakrishnan, Guru Parulkar, Larry Peterson, Jennifer Peterson, Scott Shenker, Jonathan Turner. "Openflow: Enabling Innovation in Campus Networks." ACM SIGGCOMM Computer Communication Review 38, number 2 (April 2008). [5] Nikhil Handigol, Brandon Heller, Vimalkumar Jeyakumar, Bob Lantz and Nick McKeown. “Reproducible Network Experiments Using Container-Based emulation.” CoNEXT. Dec. 2012. [6] P.Fonseca, R. Bennesby, E. Mota and A, Passito. "A replication component for resilient openflow-based networking." NOMS. 2012. [7] Radhika Niranjan Mysore, Andreas Pamboris, Nathan Farrington, Nelson Huang, Pardis Miri, Sivasankar Radhakrishnan, Vikram Subramanya, and Amin Vahdat. “PortLand: A Scalable Fault-Tolerant Layer 2 Data Center Network Fabric.” SIGCOMM. Aug. 2009. [8] Software Defined Network. https://www.opennetworking.org/sdn-resources/sdn-definition. [9] Veena S, Chandan Pal, Ram P. Rustagi, K.N.B.Murthy. “A Framework for Implementing Realistic Custom Network Topology in Mininet.” International Journal of Science and Research (IJSR), ISSN (Online):2319-7064 3, no. 7 (July 2014): 1316-1323. [10] Veena S, Chandan Pal, Ram P. Rustagi, K.N.B.Murthy. “Implementation of Simplified Custom Topology Framework in Mininet.” 2nd Asia Pacific Conference on Computer Aided System Engineering. Bali, Indonesia: ISBN 978-0-9924518-0-6, Feb. 2014. [11] Veena S, Ram P. Rustagi, K.N.B.Murthy. “Application of SDN in Network Management and Performance Minitoring.” 18th Annual International Conference IEEE ADCOM'12. Bangalore, India, Dec. 2012.