УПРАВЛЕНИЕ И ОБРАЗОВАНИЕ MANAGEMENT AND EDUCATION TOM IX (6) 2013 VOL. IX (6) 2013 POX CONTROLLER PERFORMANCE FOR OPENFLOW NETWORKS Selçuk Yazar, Erdem Uçar POX CONTROLLER ЗА OPENFLOW ПЛАТФОРМА Селчук Язар, Ердем Учар ABSTRACT: OpenFlow is an open platform that manages flow tables and packet forwarding in the network switches and other network components. The OpenFlow controller, connected network switch s control path with secure channel that is basically flow control software. This paper focused performance of newly adopted OpenFlow controller POX in a simple network controlled by a HP network switch. Jitter values analyzed with different size of packets and three distinct state of OpenFlow protocol. Keywords: OpenFlow, POX Controller, Software Defined Networking Introduction Increase of requirements of networking Technologies, concept of Software defined network(sdn) became more popular last five years. SDN, Network virtualization have attracted a lot of attention. Researches in Universities and research Institutes expect more flexible and unified network management methods and also they demand develop and test new network algorithm and Technologies. OpenFlow one of the most crucial part of SDN was developed for these purposes. One of the aims of this paper provide an numerical values using with OpenFlow protocol and POX controller for middle sized networks where installed universities and small business companies. Network flow is defined in RFC3697 briefly like a sequence of packets sent from a particular source to a particular unicast, anycast, or multicast destination that the source desires to label as a flow [1]. OpenFlow protocol is give an API that allows control the flow. OpenFlow is an open standard that enables researchers to run experimental protocols in the campus networks we use every day [2]. Concept of software defined network creates an alternative view for manageable network components. Most significant part of networks is a network switch which consist of three parts: Management Plane Forwarding Plane Control Unit Today s network switches are close boxes and researchers cannot modify this structure. With OpenFlow protocol API to control switch flow table which stored in switches Control Unit. The OpenFlow network architecture comprises several OpenFlow switches, one or more controller, a secure channel that interconnects the switch with the controller, and the openflow protocol for signaling between the switches and the controller. A proprietary protocol de nes the messages exchanged between an OpenFlow switch and an OpenFlow controller: messages include information on received packets, sent packets, forwarding table modi cations, statistics collection, etc. When the switch receives a packet that does not match any entry in the ow table, it sends the packet to the controller. The controller may drop the packet or may add a ow entry in the switch ow table, to force the switch to forward packets belonging to the same ow on a given path. Each entry in the ow table of an OpenFlow switch contains a set of packet header elds, an action to be performed on packets matching the header eld, and ow statistics. The packet header eld includes: port id; VLAN tag; Ethernet type, source and destination address; IP protocol type, source and destination address; UDP/TCP source and destination port. Examples of possible actions are: send the packet to output port i, modify packet header eld f or drop the packet. Flow statistics include number of packets and bytes for each ow, and the time since the last packet matched the ow (to ease the task of inactive ow removal).[3] 186
Network Setup This paper focused POX OpenFlow controller performance in a simple test network located disparate geographic locations with five distinct size of network packets for eight hours. In our test environment used three PC and one network switch described below, OpenFlow Controller Intel Xeon 2.67 GHz Processor 6GB of RAM Windows 7 OS Terminal 1 : Intel Core i3 Processor 4GB of RAM Windows 8 OS Terminal 2: Pentium 4 2.40 GHz Processor 128 MB RAM Windows XP SP3 Also D-ITG software used for generate network traffic packets and calculation jitter values. In summary jitter is variation of network delay[4]. Finding distribution of network latency data shown in [5]. D i = (R i S i) - (R i -1 S i - 1 ) D i = (R i R i - 1) - (S i S i - 1 ) AvgJitter = [6] Fig. 1. Network Setup 187
Test Results To see the values of all the work on the network, a statistical study was conducted on the network delay variation. Our hypothesis is here find out POX controller whether a specific effect of network switch s performance in three different cases. Evaluation of results for 64 bytes packets In this study or 64 byte of network packets, there is no significant difference with POX Controller and Switch performance. Study of Fig. 2- b and statistics of values, the difference can be seen about 15 ms. at beginning of session. Fig. 2-a. The distribution for 64 byte packets delays. Descriptive Statistics OpenFlow & Controller Status Mean Std. Deviation N OpenFlow Disabled. 19,72010252 1,320077647 484 OpenFlow Enabled No POX 19,55555622 2,256485732 484 POX 19,55710093 2,141973679 484 Total 19,61061828 1,952909405 1448 Table 2-a. 64-bytes packets statistics for the values of the Jitter. Tests of Between-Subjects Effects Source Type III Sum of Squares df Mean Square F Sig. Corrected Model 8,607 a 2 4,304 1,129,324 Intercept 556875,160 1 556875,160 146039,674,000 OpenFlow & ControllerStatus 8,607 2 4,304 1,129,324 Error 5510,041 1445 3,813 Total 562385,202 1448 Corrected Total 5518,648 1447 a. R Squared =,002 (Adjusted R Squared =,000) 188
Table 2-b. Distribution of Jitter values of 64 bytes packets. Fig. 2-b. For Jitter of 64 bytes packets measurements. Evaluation of results for 128 bytes packets When packets sizes up to 128 bytes, there is no significant difference both three distinct status of network structure. Results of Controller and OpenFlow protocol status can be seen at Fig 3-b. There is 12 ms. average of these three states at beginning of session. Fig. 3-a. The distribution for 128 bytes packets delays. Descriptive Statistics Dependent Variable: Jitter(s) OF & Controller Status Mean Std. Deviation N OF Disabled 19,85799681 2,420571452 484 OF Enabled No POX 19,87413446 2,317742501 484 POX 19,74183942 2,371526112 484 Total 19,82465690 2,369418035 1452 Table 3-a. 128-bytes packets statistics for the values of the Jitter.. 189
Tests of Between-Subjects Effects Source Type III Sum of Squares df Mean Square F Sig. Corrected Model 5,042 a 2 2,521,449,639 Intercept 570660,715 1 570660,715 101569,773,000 OpenFlow & Controller Status 5,042 2 2,521,449,639 Error 8141,077 1449 5,618 Total 578806,834 1452 Corrected Total 8146,120 1451 a. R Squared =,001 (Adjusted R Squared = -,001) Table 3-b. Distribution of Jitter values of 128 bytes packets. Fig. 3-b. For Jitter of 128 bytes packets measurements. Evaluation of results for 512 bytes packets When packet size set 512 bytes, three states of OpenFlow average jitter value became approximately 20 ms. at beginning of session. It can be seen Fig. 4.b. This value related to packet size. 190
Fig. 4-a. The distribution for 512 bytes packets delays. Descriptive Statistics OpenFlow & Controller Status Mean Std. Deviation N OpenFlow Disabled. 20,37354376 2,838713071 484 OpenFlow Enabled No POX 20,25169131 2,970903050 484 With POX 20,44275731 2,891020274 484 Total 20,35599746 2,899798243 1452 Table 4-a. 512-bytes packets statistics for the values of the Jitter. Tests of Between-Subjects Effects Source Type III Sum of Squares Df Mean Square F Sig. Corrected Model 9,058 a 2 4,529,538,584 Intercept 601660,351 1 601660,351 71505,481,000 OpenFLow & Controler Status 9,058 2 4,529,538,584 Error 12192,154 1449 8,414 Total 613861,563 1452 Corrected Total 12201,212 1451 a. R Squared =,001 (Adjusted R Squared = -,001) Table 4-b. Distribution of Jitter values of 512 bytes packets. 191
Fig. 4-b. For Jitter of 512 bytes packets measurements Evaluation of results for 1024 bytes packets In the test study for 1024 bytes packets, difference of three states of OpenFlow protocol was calculated approximately 35 ms. at beginning of session. Increasing the size of packet, such a difference is thought to occur. Results can be seen in Fig. 5-b. Fig. 5-a. The distribution for 1024 bytes packets delays. Descriptive Statistics OpenFlow & Controller Status Mean Std. Deviation N OpenFlow Disabled. 21,02074278 3,706450559 484 OpenFlow Enabled No POX 21,16146350 3,582484118 484 With POX 21,39098719 3,736156773 484 Total 21,19106449 3,676269509 1452 Table 5-a. 1024-bytes packets statistics for the values of the Jitter. 192
Tests of Between-Subjects Effects Source Type III Sum of Squares df Mean Square F Sig. Corrected Model 33,810 a 2 16,905 1,251,286 Intercept 652036,883 1 652036,883 48262,283,000 OpenFlow & Control Status 33,810 2 16,905 1,251,286 Error 19576,394 1449 13,510 Total 671647,086 1452 Corrected Total 19610,203 1451 a. R Squared =,002 (Adjusted R Squared =,000) Table 5-b. Distribution of Jitter values of 1024 byte packets. Fig. 5-b. For Jitter of 1024 bytes packets measurements Evaluation of results for 2048 bytes packets Finally for 2048 bytes packets average value of three states of OpenFlow is approximately 37 ms. at beginning of session. This value very close to 1024 bytes packets. 193
Fig. 6-a. The distribution for 2048 bytes packets delays. Descriptive Statistics OpenFlow & Controller Status Mean Std. Deviation N OpenFlow Disabled 22,02414475 4,775860857 484 OpenFlow Enabled No POX 22,27211203 4,582169985 484 With POX 22,06652740 4,798145473 484 Total 22,12092806 4,717712152 1452 Table 6-a. 2048-byte packets statistics for the values of the Jitter Tests of Between-Subjects Effects Source Type III Sum of Squares df Mean Square F Sig. Corrected Model 17,029 a 2 8,514,382,682 Intercept 710515,085 1 710515,085 31896,311,000 OpenFlow & Controler Status 17,029 2 8,514,382,682 Error 32277,600 1449 22,276 Total 742809,714 1452 Corrected Total 32294,628 1451 a. R Squared =,001 (Adjusted R Squared = -,001) Table 6-b. Distribution of Jitter values of 2048 bytes packets. 194
Fig. 6-b. For Jitter of 2048 bytes packets measurements Conclusion All results were evaluated, no significant changes was observed for POX controller and network switch whether OpenFlow protocol enabled or not. Considering the differences POX Controller is thought to be preferable software for small and medium businesses Networks. Because of POX Controller developed with python language, it offers platform independent choice. POX controller could work environments where Python runtime could work, so this feature of python give researches new areas form mobile systems to embedded systems. Abbreviations POX : Pythonic Network Operating System SDN : Software Defined Networking VLAN : Virtual Local Area Network API : Application Program Interface TCP : Transmission Control Protocol UDP: User Datagram Protocol IP : Internet Protocol VPN : Virtual Private Network References 1. N. Brownlee, C. Mills, and G. Ruth. RFC 3697 - IPv6 Flow Label Specification. IETF. (2004).pp.1. 2. N. McKeown, T. Anderson, H. Balakrishnan, G. Parulkar, L. Peterson, J.Rexford, S.Shenker, J.Turner. OpenFlow: Enabling Innovation in Campus Networks.(2008). U.S.A. 3. A. Bianco, R. Birke, L. Giraudo, and M. Palacin. OpenFlow Switching: Data Plane Performance. IEEE ICC. (2010).pp.1. 4. Tim Szigeti,Kevin McMenamy,Roland Saville. Cisco TelePresence Fundamentals. Cisco Press.(2009).pp.79. 5. Stefano Avallone, Alessio Botta, Alberto Dainotti, Walter de Donato, and Antonio Pescap e. D-ITG V. 2.7.0-Beta2 Manual. University of Napoli Federico II..Italy.(2009). Selçuk YAZAR Trakya University Computer Engineering Trakya Üniversitesi Balkan Yerleşkesi 22030 Edirne/Turkey selcukyazar@trakya.edu.tr 195