FlowPing - The New Tool For Throughput And Stress Testing

INFORMATION AND COMMUNICATION TECHNOLOGIES AND SERVICESVOLUME: 13 NUMBER: 5 2015 DECEMBERFlowPing - The New Tool for Throughput andStress TestingOndrej VONDROUS, Peter MACEJKO, Zbynek KOCURDepartment of Telecommunication Engineering, Faculty of Electrical Engineering,Czech Technical University in Prague, Technicka 2, 166 27 Prague, Czech Republicondrej.vondrous@fel.cvut.cz, peter.macejko@fel.cvut.cz, zbynek.kocur@fel.cvut.czDOI: 10.15598/aeee.v13i5.1497Abstract. This article presents a new tool for networkthroughput and stress testing. The FlowPing tool iseasy to use, and its basic output is very similar to standard Linux ping application. The FlowPing tool is notlimited to reach-ability or round trip time testing but iscapable of complex UDP based throughput stress testing with rich reporting capabilities on client and serversides. Our new tool implements features, which allowthe user to perform tests with variable packet size andtraffic rate. All these features can be used in one singletest run. This allows the user to use and develop newmethodologies for network throughput and stress testing. With the FlowPing tool, it is easy to perform thetest with the slowly increasing the amount of networktraffic and monitor the behavior of network when thecongestion occurs.Flowping tool utilize UDP (User Datagram Protocol) [2]. One of the reason to use UDP protocol is thatit is not a priory blocked or influenced by traffic controlpolicies on gateways and firewalls like standard pingapplication which utilize ICMP (Internet Control Message Protocol) protocol [3]. Another reason for usingUDP protocol is the fact that our other application [4]uses UDP as the underlying protocol for communication and thus we can easily compare these applicationmeasurement results with FlowPing results.The main reason for developing and publishing theFlowPing tool under open source license is the fact thatthere are many open source tools such as Iperf2, iperf3,Ostinato, Seagull, pacgen, Bittwist, Nping, CLAudit[5] and others. Appointed tools allow the user to generate general traffic flow to stress test network, but wedo not know any open source tool, which is capableof generating variable increasing or decreasing trafficKeywordsflows. In addition, our tool has some unique featuressuch as high precision time reporting. On the otherPing, stress testing, throughput, variable traffic hand, there are commercial tools (such as Ixia IxChariot) but they are too expensive for general use. That israte.also why we decided to provide this tool free of chargeunder Creative Commons 3.0 BY-NC-SA License.1.IntroductionIn this article, we would like to introduce the new toolfor network throughput and stress testing. We nameour new tool FlowPing [1].2.FlowPing ToolThis tool is aimed to be used especially for latency andThe main difference between our tool and other com- network throughput stress testing. The FlowPing toolmonly used open source tools is that our tool allows us can be run on several platforms like x86-32, x86-64 andto examine the behavior of networks in reaction to the ARM.dynamic change of generated traffic amount.This tool is able to send and to receive UDP packetsThe unique feature of our tool is the ability to gener- in the very similar way as the standard ping applicationate increasing or decreasing traffic data flow. The tool handles ICMP packets. The basic output of this toolis capable of generating complex variable traffic flows is similar to well known ping application. That is whybecause it is possible to read complex test scenario from the usage of FlowPing tool is very intuitive. On theother hand this tool has more features than standarda file.c 2015 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING516

INFORMATION AND COMMUNICATION TECHNOLOGIES AND SERVICESping tool and it is not limited only for network reachability testing but it covers large variety of networkthroughput and stress tests.The great advantage of the FlowPing tool is the possibility of writing prescription of test into file. This isplain text file (it is possible to use white spaces, comas or semicolons as the field separators) with simpleand intuitive structure (see Fig. 1). It allows users todefine very complex tests therefore user can save largeamount of time when performing network stress tests.That is due to the fact that many different tests can becombined into one complex test and it is not necessaryto run every single test manually.# t 512128128Packet 000VOLUME: 13 NUMBER: 5 2015 DECEMBERThe FlowPing tool have many other interesting features as described on applications homepage [1].3.Traffic GeneratorPerformance and PrecisionThe FlowPing tool is capable of running tests with variable inter-packet intervals resulting in a variable rateof data flow (increasing or decreasing). The tests withvariable packet size are also available. Figure 2 showsthese possibilities of variable flow definition includingpacket size. It is also possible to define the variable flowtest on the command line (only simplified scenario).The FlowPing tool allows user to use special modesto increase traffic generator accuracy. It is possible toselect passive waiting mode which is effective in wayof CPU utilization, but with lower accuracy (similarto Iperf 2 tool). If higher accuracy is important it ispossible to use busy loop mode (active waiting) whichutilizes at least 100 % of one CPU core for duration oftest. The traffic generator stability and accuracy aregreatly increased in such case.The FlowPing tool also implements the possibility to compute all packet intervals before test starts.This approach combined with busy loop mode greatlyincreases packet timing precision and overall performance. This feature is internally associated with otherfeature which stores all output in memory and writesresults after the test is finished. These features ensureFig. 1: FlowPing - test configuration file data structure.maximum possible performance and timing precisionof FlowPing tool. It is possible to generate traffic ratesThe extended output can provide advanced statis- up to 1 Gbps on standard Linux machine (applicationtics such as immediate sending and receiving bit rate, runs in user space on Intel i5-2500k with packet size ofinter-packet intervals and time stamps with nanosec- 1470 B).ond resolution. The FlowPing tool is also capable ofWe have performed many tests to study the Flowstoring and displaying statistics in CSV format for furPingtool performance. We have mainly focused onther data processing.traffic generator precision and stability. The FlowPingtool is optimized to achieve comparative or even betterresults than well-known and widely used tools such as102464 Bthe Iperf 2 [6] or the iperf 3 [7].500 B24 B1472 B1000 B896TX Rate [kbps]768The results of traffic flow generator stability test(test setting were as follows: Flowping [-b 5000 -s 160],Iperf 2 & 3 [-u -b 5M -l 160] ) of FlowPing, Iperf 2and iperf 3 tools are shown in Fig. 3. From top leftto bottom right there is comparison of packet delaystability of Flowping tool in standard mode, FlowPingtool with packets interval computed before test start(utilizing busy-loop waiting), Iperf tool version 2 andfinally totally redesigned iperf at version 3.64051238425612800153045607590 105Time [s]Fig. 2: FlowPing - variable traffic flow.120135150175180For validation purposes we used tcpdump utility tocapture packets directly on outgoing interface to ensureobjectivity of measurements when dealing with trafficgenerator stability tests. This approach allows us toc 2015 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING517

INFORMATION AND COMMUNICATION TECHNOLOGIES AND SERVICESVOLUME: 13 NUMBER: 5 2015 DECEMBER10Packet No. (x103)Packet No. (x103)compare results from FlowPing tool with results from to observe dynamic network behavior in reaction tosmall traffic amount changes. This approach is veryother tools such as Iperf version 2 or iperf version 3.useful for observing UE (User Equipment) resource al2020FP -W (busy loop)location in mobile wireless networks as a reaction toFP (pasive waiting)1515traffic amount change.105Packet No. (x103)Packet No. (x103)With FlowPing tool, it is not necessary to repeatmeasurement with different settings again and again00because this tool allows user to create complex mea200220240260280300200220240260280300Delay [us]Delay [us]surement scenario and put it into single configuration2020file. When finding congestion point of network, espeiPerfiPerf31515cially this method is very effective. Utilizing variableflows for network stress testing can be very effective in1010situation when it is necessary to find the exact amount55of traffic which causes network congestion. In first run,002002202402602803000500 1000 1500 2000it is possible to perform very quick test and find apDelay [us]Delay [us]proximately the traffic amount which causes congestion and in second run it is possible to target the rangeFig. 3: Packet delay stability - FlowPing vs. Iperf 2 vs. iperf 3.of test on this congestion point and perform precisemeasurements just around this congestion point of theThe results show that the precision of traffic gener- network.ator is mainly determined by waiting mode used. ReIn following sections we would like to present severalsults also showed that traffic generator was completelycases when usage of variable flow in stress testing isredesigned in iperf3. Iperf3 generates packets in bursts.useful.At first it generates packets at maximum speed for thespecific time period and then it adds longer delay tocreate average data rate.54.1.Congestion Point Detection200150100RTT [ms]The variable flows can be very valuable source ofdata for network parameter estimation. It is possibleto use it for observing the dynamic behavior of networkunder changing load. It is also possible to detect trafficengineering methods such as network buffering alongnetwork path, traffic shaping or traffic policing.TX/RX rate [kbps]The traffic generator precision is also influenced bycurrent system load of machine where the FlowPingThis test is used to find the exact amount of traffictool is running.flow when congestion occurs. It is very convenient touse this type of network throughput test in a situation when the parameters of the network are unknown.4.Variable Flow StressOn the chart in the Fig. 4 is shown the difference between generated traffic flow and received traffic flow.TestingBy comparing sending and receiving traffic, RTT andloss rate you can very easily find exact throughput andWe found feature of generating variable traffic flow esthe point where congestion occurs. It is also possiblesential for measurement of important network paramto detect if some mechanism of traffic control was usedeters.as shown in the subsection 4.2.Methods utilizing variable traffic rates can simplifynetwork testing in some cases. The measurements uti350250Client TXlizing variable data rates make detection of traffic shapServer RX300Client-Server RTTing and policing easier as opposed to methods used in200article: "End-to-end detection of isp traffic shaping us250ing active methods" [8] where sustained date rate wasused for active network detection of traffic policing.15010050500Methods for network throughput and stress testing020406080100based on increasing and decreasing the amount of netTime [s]work traffic provide valuable information about networks under changing load. This approach allows us Fig. 4: Congestion point detection.c 2015 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING1201401600180518