高级计算机网络实验报告 ns3模拟数据中心

1、Project1-ns3模拟数据中心实验要求Datacenter network topology1根据上面的数据中心拓扑图，完成以下要求:根据给定的数据中心的拓扑结构，利用ns3进行仿真模拟两种通信模式(traffic pattern)o all-to-all :每个服务器都发送消息给其他服务器消息，由拓扑结构可知，超过50%的消息传送将跨越两个簇(cluster)o many-to-one :每个服务器都发送消息给其中一个服务器测量两种模式下网络的仿真可以达到的吞吐量，找出网络瓶颈，并且说明如何改进注：拓扑中的网络都是Ethernet网实验内容数据中心模拟实现及主要代码解释a.设置自定义的

2、attribute为了做实验方便，设置如下自定义attribute :pattern:通信模式，all-to-all 或 many-to-one，默认为 1defaultDst:多对一模式下，接收消息的默认服务器序号，默认为0verbose:enable 或者 disable PacketSink和 OnOffApplication 的日志，默认为 falseDataRatel:定义数据中心拓扑第一层的数据传输速率(Mbps)，默认为1.0DataRate2:定义数据中心拓扑第二层的数据传输速率(Mbps)，默认为1.0DataRate3:定义数据中心拓扑第三层的数据传输速率(Mbps)，默认

3、为1.5实现代码如下：uint16_t pattern = 1;uint16_t nodesNum = 8;uint16_t defaultDst = 0;float DataRate1 = 1.0;float DataRate2 = 1.0;float DataRate3 =1.5;uint16_t port = 50000;bool verbose = false;CommandLine cmd;cmd.AddValue(pattern， number of traffic pattern, pattern);/pattern1:all-to-all pattern2:many-to-on

4、ecmd.AddValue(defaultDst, default destination server node in pattern 2, defaultDst);cmd.AddValue(DataRate1， data rate of csma network at level 1, DataRatel);cmd.AddValue(DataRate2, data rate of csma network at level 2, DataRate2);cmd.AddValue(DataRate3, data rate of csma network at level 3, DataRate

5、3);cmd.AddValue (verbose, Tell sink and onoff applications to log if true, verbose);cmd.Parse(argc, argv);LogComponentEnable (DataCenterSimulation, LOG_LEVEL_INFO);if (verbose)(LogComponentEnable (PacketSink, LOG_LEVEL_INFO);LogComponentEnable (OnOffApplication, LOG_LEVEL_INFO);b.创建结点根据实验要求，总共需要创建15

6、个结点，包括：8 servers4 ToR switches2 Aggregation switches1 Core switch实现代码如下：/create nodesNodeContainer n1_8;n1_8.Create(8);NodeContainer t1_4;t1_4.Create(4);NodeContainer a12;a12.Create(2);NodeContainer c1;c1.Create(1);c.创建CSMA网络节点整个数据中心网络拓扑从下往上可以分为三层，即第一层：由服务器与ToR组成的ethernet网络，共有4个，编号为CSMA11,CSMA12,CSM

7、A13,CSMA14第二层：由ToR与Aggregation组成的ethernet网络，共有2个，编号为CSMA21,CSMA22 第三层：由Aggregation与Core组成的ethernet网络，共有1个，编号为CSMA3将创建好的15个网络结点分配到这7个CSMA网络中，实现代码如下：/create csma nodesNodeContainer csmaNodes11 =NodeContainer(n1_8.Get(0),n1_8.Get(1),t1_4.Get(0);NodeContainer csmaNodes12 =NodeContainer(n1_8.Get(2),n1_8.

8、Get(3),t1_4.Get(1);NodeContainer csmaNodes13 =NodeContainer(n1_8.Get(4),n1_8.Get(5),t1_4.Get(2);NodeContainer csmaNodes14 =NodeContainer(n1_8.Get(6),n1_8.Get(7),t1_4.Get(3);NodeContainer csmaNodes21 =NodeContainer(t1_4.Get(0),t1_4.Get(1),a12.Get(0);NodeContainer csmaNodes22 =NodeContainer(t1_4.Get(2

9、),t1_4.Get(3),a12.Get(1);NodeContainer csmaNodes3 =NodeContainer(a12.Get(0),a12.Get(1),c1.Get(0);d.设置CSMA网络attribute，并将其安装到相应结点上根据实验要求中的网络拓扑，设置相应网络的属性所有直接相连的两个结点之间的延迟都为500ns第一层和第二层CSMA网络的数据传输速率都为1.0Mbps，第三层为1.5Mbps然后安装到相应的网络结点上，实现代码如下（DataRate可以通过命令行参数设置，默 认值即为原实验要求）：/create the channels first witho

10、ut any IP addressing information CsmaHelper csma1;sprintf(buf, 1.1fMbps,DataRate1);csma1.SetChannelAttribute (DataRate”, StringValue (buf);csma1.SetChannelAttribute (Delay, StringValue (500ns);NetDeviceContainer csmaDevices11 = csma1.Install (csmaNodes11);NetDeviceContainer csmaDevices12 = csma1.Ins

11、tall (csmaNodes12);NetDeviceContainer csmaDevices13 = csma1.Install (csmaNodes13);NetDeviceContainer csmaDevices14 = csma1.Install (csmaNodes14);CsmaHelper csma2;sprintf(buf, %1.1fMbps”,DataRate2);csma2.SetChannelAttribute (DataRate, StringValue (buf);csma2.SetChannelAttribute (Delay, StringValue (5

12、00ns);NetDeviceContainer csmaDevices21 = csma2.Install (csmaNodes21);NetDeviceContainer csmaDevices22 = csma2.Install (csmaNodes22);CsmaHelper csma3;sprintf(buf, %1.1fMbps”,DataRate3);csma3.SetChannelAttribute (DataRate, StringValue (buf);csma3.SetChannelAttribute (Delay, StringValue (500ns);NetDevi

13、ceContainer csmaDevices3 = csma3.Install (csmaNodes3);根据实验要求，为每个结点安装协议栈，并为7个CSMA网络分配IP，实现代码如下/assign IP addressNS_LOG_INFO (Assign IP address.);InternetStackHelper stack;stack.Install (n1_8);stack.Install (t1_4);stack.Install (a12);stack.Install (c1);Ipv4AddressHelper address;address.SetBase (10.0.1

14、.0, ”255.255.255.0);Ipv4InterfaceContainer csmaInterfaces11 = address.Assign(csmaDevices11);address.SetBase (10.0.2.0, ”255.255.255.0);Ipv4InterfaceContainer csmaInterfaces12 = address.Assign (csmaDevices12);address.SetBase (10.0.3.0, 255.255.255.0);Ipv4InterfaceContainer csmaInterfaces13 = address.

15、Assign (csmaDevices13);address.SetBase (10.0.4.0, 255.255.255.0);Ipv4InterfaceContainer csmaInterfaces14 = address.Assign (csmaDevices14);address.SetBase (10.1.1.0, 255.255.255.0);Ipv4InterfaceContainer csmaInterfaces21 = address.Assign (csmaDevices21);address.SetBase (10.2.1.0, 255.255.255.0);Ipv4I

16、nterfaceContainer csmaInterfaces22 = address.Assign (csmaDevices22);address.SetBase (192.168.1.0, 255.255.255.0);Ipv4InterfaceContainer csmaInterfaces3 = address.Assign (csmaDevices3);f.初始化路由表这里直接调用了 ns3自带的路由实现，实现代码如下/ Create router nodes, initialize routing database and set up the routing/ tables i

17、n the nodes.Ipv4GlobalRoutingHelper:PopulateRoutingTables ();g.创建和分配 PacketSink 和0nOffClient首先，创建sink和OnOff，实现代码如下/Create sinkApp and OnOffClientApplicationContainer clientAppnodesNum4;ApplicationContainer sinkAppnodesNum;然后，分配sink到所有的server结点上，实现代码如下俱中nodesNum表示server个数):for (unsigned int i = 0;i n

18、odesNum; i+)(PacketSinkHelper packetSinkHelper (ns3:TcpSocketFactory， getAddress(i,port,csmaInterfaces11,csmaInterfaces12,csmaInterfaces13, csmaInterfaces14);sinkAppi = packetSinkHelper.Install (n1_8.Get (i);sinkAppi.Start(Seconds (1.0);sinkAppi.Stop(Seconds (60.0);再然后，分配OnOffClient到server结点上，并且根据pa

19、ttern不同，进行不同的配置pattern 1 ：每个服务器都发送消息给其他服务器消息，即发送消息给在另一个簇上面的4 个服务器(每个服务器上建立4个OnOffClient)pattern 2 ：每个服务器都发送消息给同一个服务器，可以默认为n1(每个服务器(n1除外)上建立1个OnOffClient)实现代码如下for (int i = 0; i nodesNum; i+)(uint16_t dst = 0;if (pattern=1)(/all-to-all patternfor(int j = 0 ;j 4; j+)(if(i nodes- chanfi&ls. IP address.

20、 sinkApp and dniDfFClient-1 build Create Create/ns-alLLnorie-l. 1 g / n & n-a-alltnofte-i. iB/ihs -run scratch/DC-3.is/bulld,-patterfi=lCreatefClient feiient pnDff Cl-Lent WftOffGlicnt pOffClicnt DoOffCllent hnOffClient OoOffClLtenit OnOffClient hoOffClient jOnOff Client OnOffClient inQf fcltent fif

21、tOff Client (iDffClLlent jo naff Client OhOffClient OftOf fCltent OnOf fdtent OftOf fell tent flftOffClient OmDfrcilent CiiiOfrcitent OftorrcitentNodc Node Node Mode E*Eck gNGdc gfriodc iNode gwodr gfiMQdc gwodegwode *Jode gfrJode 整Mode* gfMode gfM&de Node gMode 耕1凸施 glMode godeains aifis-auns-flitf

22、lS ai nsot*is ains aiis otfis aifis atfis atns ainssins806&atns ai/is 巩/is at/is atfis atnsatatat atat at at at at at at at at at at at at at at at at atServer Server Server Server Strvtr Server Server Serwer Serwer Server Server Server Server Server Server Server Server Server Server Server Server

23、Server Server ServerNlade Nnde hiadeflNode gpdu Mode 0Mode ModeNode Mode Mode Mode NodeMode (pHode 前白血 0Nioie 前口 de* $Nsode iNcideTCP Stream Graph-Throughput Graphs看整个过程中该结点上的吞吐量变化情况使用statistics-Summary来查看当前结点上网络平均吞吐量，以此估计相应CSMA网络的吞吐量server n1上的测量结果如下Throughput GraphThroughputThFoughpMt Cr 叩 hdSMM ?

24、(KXX)-TraFficCapturedDisplayedMarkedPackets413941330Between Firit and lasL packet转.973 sec48.972 secAvg. packels/s&tS4.51684.355Avg. packet size377.211 bytes377. kbytesBytes15612761560892Avg. bytes/sec31830.23531S73.061Zwg, MBit/sec0.2550.255ToR t1上的测量结果如下Throughput GraphTCP Graph 1; DataCentersimul

25、ition-B-1Ppeap 1 1,1.1249153 - 1040,3,1;50000Throughput GraphThrough|utB/i7Q0M 60000 400M 汹的一100K 20304045Times|TraFficCapturedDisplayedMarkedPackets8604859&0Between First and hst packet4S.986 5ee48.985 SAvg. packets/sec175-643175.525Avg. packet size384.747 bytes384.971 bytesBytes33103663309982Avg.

26、bytes/sec6757S.36267571.938Avg. MBit/sec0.5410.541Aggregation al上的测量结果如下ThroughplitTrafficCapturedDisplayedMarkedPacketetweer first and last packet48.990 sec48.989 SECAvg. packets/sec323.476323.399Avg. packet size383.565 bytes3 S3.646 bytesBytes6078358607S10ZAvg. bytes/sec124074.16&1240

27、70.677Avg. MBit/s0.9930.993b. pattern 1实验结果分析首先对实验结果进行简单汇总网络结点带宽(Mbps)网络平 均gUB(Mbps)CSMA11server n11.00.255CSMA21ToR t11.00.541CSMA3Aggregation a11.50.993从上面的结果可以看出第1层CSMA网络平均吞吐量是0.255Mbps，带宽利用率为25.5% ;第2层CSMA网络平均吞吐量是0.541Mbps，带宽利用率为54.1% ；第3层CSMA网络平均吞吐量是0.993Mbps，带宽利用率为662%。c. pattern 1瓶颈及改进瓶颈根据以上的

28、实验结果可以看出来，从网络的平均吞吐量来看：第3层CSMA第2层CSMA第1层CSMA2，从带宽利用率上看也是这样，所以作为core switch连接两个子 网络但带宽过小的第三层网络成了整个网络的瓶颈。改进可以加大第3层网络的带宽，防止数据流量过大出现拥塞的情况发生。因此最后确定的网 络带宽如下所示：o CSMA11-14 : 1.0Mbpso CSMA21-22 : 1.0Mbpso CSMA3 : 2.0Mbps改进结果执行命令彳亍/waf -run scratch/DC -DataRate1=1.0 -DataRate2=1.0 -DataRate3=2.0，以相同方式测量网络，得到的

29、结果如下：网络结点带宽(Mbps)网络平 均SUB(Mbps)CSMA11server n11.00.306CSMA21ToR t11.00.523CSMA3Aggregation a12.00.990从上面的结果可以看出加大第三层网络的带宽，确实提高了整个网络的吞吐量，尤其是对 最底层的server结点来说。d. pattern 2实验结果执行命令./waf -run scratch/DC -pattern=2，实验运行结果如下所示atjisaxns8N0de3.185PCDPpspNode N&de gMode BN&de griodeServer Server Server SerwrS

30、erver Server Server ServerbuildCreateCreateAssignCreateOflorrciient aifisstnkApp ancrffclient onoffcllent nDffEllent OhDffClUnt OnOffClient oniJffclient OnOffClient anDrffclient Slnrulator beg Ln. Silator done,.Uiur zimu r z vt r tu * iBox t - /kJor k? pace/advfie-t / pro J1 /ns - a l L1 n 3,18 / n

31、s - i, 185 , /w*T - r Datacenterstnulattoo-9-0. Detacent已rslEulatlGn9-I*/示从上面的结果可以看出，8个OnOffClient都将数据传输到n1，产生的 pcap 文件如下所为了分析Patten 2的网络吞吐量根据网络拓扑的对称性，选取测量位置 CSMA11 :通过 server n1(node 0/device 0)来估计 CSMA12 :通过 server nnode 2/device 0)来估计CSMA13 :通过 server n5；node 4/device 0)来估计CSMA21 :通过 ToR t1(node

32、8/device 1)来估计CSMA22 :通过 ToR t3(node 10/device 1)来估计CSMA3 :通过 Aggregation a1node 12/device 1)来估计选取的测量标准为吞吐量，具体方法是使用wireshark分析相应的pcap文件中的tcp包 使用 statistics-TCP Stream Graph-Throughput Graphs看整个过程中该结点上的吞吐量变化情况使用statistics-Summary来查看当前结点上网络平均吞吐量，以此估计相应CSMA网络的吞吐量server n1上的测量结果如下Throughput GraphThrough

33、putTCP Grap80000 70000 MOOCi 50000 40000 30000 20000 10000 TraFHcCapturedDisplayedMarkedPackets14&31148230Between first and last packet4&.9B9 sec48.9 & secAvg. packets/sec302.744302.5 &7Avg. packet size401.958 bytes402.141 bytesBytes59614465 泌0934Avg. by tes/s &c121690.443121682.542Avq. MBit/sec0.97

34、40.973server n3上的测量结果如下Throughput Graph9000G &0000-7M0D 60000-.50000 4M00 30000 2 WOO-10000TCP Graprhroijghput 网10M00 TraFfieCapturedDisplayedMarkedPackets27912785aBetween first and Last packet48.925 sec48.924 secAvg. packeti/sec57.04656.925Avg. packetsize396.318 bytes397.034 bytesBytes1106124110574

35、0Avg. bytes/sec22608.45922601.084Avq. MBit/海0.1IS10.181server n5上的测量结果如下Throughput GraphTCP Graph li DdtdCinterslimultlO.10.0.3,1:49-153 - l&.G.l-liSCKMK)Throughput GraphThroughoutB/5TQMO 60000 50000 4O0OQ 30000 20MQ 15M。TrafficCapturedDisplayedMarkedPackets179017840Between first and last packet48.9

36、35 sec48.934 secAvg. patkels/sec36.57936.457Avg. packet size392.955 byLes394.062 bytesBytes703390703006Avg. bytes/sec14373.86614366.320Avq. MBiL/sec0.1150.115ToR t1上的测量结果如下SOOM 70000 60000-EOOM 40000 30000 Througihput e/5 90000 W 1520 Z5 30354045Time520000 TrafficCapturedDisplayedMarkedPackets930492

37、980Between first and hst packet48.989 sec48.9 BO secAvg. packets/sec1&9.921189.&32Avg. packet size395.877 bytes396.091 bytesBytes36S32363682852Avg. bytes/sec.75185.40675190.755Avg. MBi t/sec0.6010.602ToR t3上的测量结果如下Throughput GraphTCP Graph 1： OtaCenterSinriLlIatiQn-lO-O.pcap 10.03.1：4915B - 10,0.1.1

38、:50000ThroughputB/s|thro ugh put Grap h70000 600M 30000-354045Time400W 200M TrafficCapturedDisplayedMarkedPackets275427460Between first and last packet4&.940 sec48.939 secAvg. packets/sec56.27356.111Avg. packet size395.587 bytes396.553 bytesBytes10894461088934Avg. by tes/s &c22261.04222251.047Avq. MBit/sec0.1780.117&Aggregation al上的测量结果如下Throughput Graphtcf Craph i： Oataccnterimulati?nlz-a.pcdp 10,0.4,1:415 -a 10,0.l-i：500oaThroughput Grep-hThroughput B/s90000 80000-70000 soooo50000-40000 30000-20000 TraFficCapturedDisplayedMarkedPackets557455680Between first and