Network Troubleshooting

Transcription

1 Network Troubleshooting by Othmar Kyas 18 Testing Network Performance An Agilent Technologies Publication Agilent Technologies

2 Testing 18 Network Performance It is a capital mistake to theorize before one has data. SIR ARTHUR CONAN DOYLE Unsatisfactory network performance is a major problem that occurs continually in data networks. Because the causes of this condition can vary widely, it is very important to supplement such terms as a slow network, a slow server, or long response times with objective measurements. The two crucial parameters to be determined are data throughput and response time. The first step in determining the network performance is to analyze the structure of the applications in use. Application architectures can be divided into three main categories: Peer-to-peer (Level 1) Client-server (Level 2) Client to primary application server and secondary data/application servers (Level 3) While no dedicated server is involved in peer-to-peer communication (for example, file transfer between two network nodes), a Level 3 application architecture involves not only the client and a primary application server, but also other servers that may function as database or application servers to the primary server. Typical Level 3 application architectures include application or Web servers with back-end connections to database servers. Response Time Measurements Once the application architecture and the network topology have been analyzed, the first step in determining and evaluating response times should be to calculate the theoretical communication delay. This figure can then be used to estimate the expected response times. Typical values for one-way transmission delays over WAN links are best obtained from the service provider or tested using a protocol analyzer. Typical values for Frame Relay links are between 30 and 60 ms. For ATM (STM-1), values of 6 µs/km and 2 ms/switch can be assumed. For a line length of 1,000 km with five ATM switches, this yields a transmission delay of 6.01 ms.

3 626 TESTING NETWORK PERFORMANCE 6 O F A B E K = A = O F A H6OI FA AC AB J E K = A = O F A H I A C =?? E C J I J = I =?? H@ E C J I = H@ I - J D A H A J * = I A # $ # I - J D A H A J * = I A ' # I - J D A H A J * H $ " I - J D A H A J * = I A. 2 # I - J D A H A J. 1 4 # I - J D A H A J * = I A # " I - J D A H A J * = I A 6 I - J D A H A J * = I A. * I - J D A H A J * = I A. I - J D A H A J * = I A 6 " # % I - J D A H A J * = I A 6 : # # $ I - J D A H A J * = I A. : " I - J D A H A J ) 7 1 # % I - J D A H A J 1 1 # I 6 O F E? A = O.,, 1 `! I 5 M EJ? D > EJ I - J D A H A J! I 5 M EJ? D 6 A 4 E C! I 5 M EJ? D 1 I 5 M EJ? D * = I A 6 I 4 K J A H # ` I Figure 18.1 Transmission delays in LAN topologies The actual response time measurements are made, in the simplest case, using Internet Control Message Protocol (ICMP) ping packets. Ping uses the ICMP Echo command to send a packet to the destination and back again. In addition to the response time, the ping command also counts the number of intermediate stations (hops) through which the packet is routed. Each hop reduces the packet s Time to Live (TTL) counter by one. If the initial TTL value is known, subtracting the final TTL yields the number of hops along the path. Initial TTL values are: Windows 3.x/95 32 Windows 98/2000/NT Routers 255 UNIX 64 or 255 C:\WINDOWS>ping Pinging with 32 bytes of data: Reply from : bytes=32 time=54 ms TTL=120 Reply from : bytes=32 time=51 ms TTL=120 Reply from : bytes=32 time=53 ms TTL=120 Reply from : bytes=32 time=54 ms TTL=120 Ping statistics for : Packets: Sent = 4, Received = 4, Lost = 0 (0% loss), Approximate round trip times in milli-seconds: Minimum = 51 ms, Maximum = 54 ms, Average = 53 ms Figure 18.2 Results of a response time test using the ping command

4 627 TESTING NETWORK PERFORMANCE Figure 18.2 shows the results of such a response time test. The measured response time varies within a relatively narrow range, from 51 to 54 ms. This means that the communication links involved in this example are not under excessive loads and are able to provide consistent response times. High variations in response time indicates that at least one network segment is under a heavy load. To determine the delay values for individual segments of the network connection, the tracert command (in UNIX: traceroute) can be used. The traceroute command performs a series of ICMP Echo requests, beginning with a TTL value of 1. Each request is repeated three times, then the TTL value is increased by one, and so on until the TTL is sufficient for the packet to reach its destination. The output of the traceroute command displays the response time for each of the three echo packets with each TTL value (see Figure 18.3). 1 <10 ms <10 ms 10 ms center1-west-wien.chello.at [ ] 2 10 ms 10 ms <10 ms rt113.chello.at [ ] 3 <10 ms 10 ms 10 ms vienna-bgp1-fe aorta.com [ ] 4 40 ms 10 ms 10 ms tk-uni-eb3-pos aorta.com [ ] 5 50 ms 60 ms 60 ms atvie202-ta.ebone.network [ ] 6 60 ms 50 ms 61 ms demun701-tb-p0-3.ebone.network [ ] 7 50 ms 60 ms 60 ms frpar601-tb-p0-2.ebone.network [ ] 8 50 ms 60 ms 60 ms frpar205-tc-p9-0.ebone.network [ ] 9 60 ms 80 ms 80 ms gblon303-tc-p1-0.ebone.network [ ] ms 70 ms 60 ms gblon304-tb-p2-0.ebone.network [ ] ms 140 ms 150 ms usnyk106-tc-p0-2.ebone.network [ ] ms 140 ms 140 ms uspen201-tb-p1-3.ebone.network [ ] ms 150 ms icm-bb11-pen-3-0.icp.network [ ] ms icm-bb10-pen-9-0.icp.network [ ] ms 141 ms icm-bb4-pen icp.network [ ] ms 161 ms 140 ms sprint-nap.cerf.network [ ] ms 141 ms 180 ms pos m.phl-bb2.cerf.network [ ] ms 161 ms 170 ms ser3-5-45m.chi-bb3.cerf.network [ ] ms pos m.sfo-bb4.cerf.network [ ] ms 210 ms 200 ms pos m.sfo-bb3.cerf.network [ ] ms 230 ms pos m.sea-bb2.cerf.network [ ] ms 221 ms pos m.sea-bb1.cerf.network [ ] ms 231 ms 220 ms boeing-gw.sea-bb1.cerf.network [ ] ms 220 ms 220 ms ms 230 ms www1.boeing.com [ ] ms 230 ms www1.boeing.com [ ] Figure 18.3 Results of a response time test using the TRACERT command

5 628 TESTING NETWORK PERFORMANCE The drawback of response time testing by the ping and traceroute methods is that the server s processing time is reflected along with the network transmission delay. The transmission delay caused by the network alone can be measured using two protocol analyzers or network probes. Response Time Measurements Using Two Protocol Analyzers Place one analyzer on either side of the network or network component to be measured. Then send a ping across the network (or component) under test. When the ping passes analyzer A, the analyzer records the arrival time A 1. When it passes analyzer B, this analyzer records arrival time B 1. When the echo response packet returns, it first passes analyzer B, which records arrival time B 2, then analyzer A, which records arrival time A 2. The latency of the device can now be calculated as ((A 2 A 1 ) (B 2 B 1 ))/2. 2 E C 6 ) 2 E C 6 * 6 ) 6 * 2 E C 4 A F O 2 E C 4 A F A 2 H J? ) = O A H ) A J M H A H 6 A I J 2 H J? ) = O A H A 6 ) ` 6 ) ` 6 * ` 6 * A M = O, A = O Figure 18.4 Network transfer delay measurements using two protocol analyzers Throughput Measurements Like response time measurements, throughput tests should also be preceded by a calculation of the theoretical maximum frame rate and throughput. Note that the throughput over a connection path can never be higher than the throughput over the link with the lowest capacity. Actual throughput is best measured by means of an File Transfer Protocol (FTP) or Server Message Block Protocol (SMB) file transfer between two stations whose data traffic is being traced with a protocol analyzer. The packets transfered should be close to the maximum size of the transport network, called the

6 629 TESTING NETWORK PERFORMANCE Maximum Transfer Unit or MTU. The average throughput of forty or fifty successive packets can be measured to yield a distribution of the connection s throughput capacity over time. An important factor in analyzing throughput is the client system. Often the client s performance has a significant influence on the measured throughput. Use a protocol analyzer to measure the time between the server s last response packet and the client s next request packet to obtain an indication of the client s speed in processing the server s responses, which has an effect on throughput as well as response time. Figures 18.5a and 18.5b show the calculation of the theoretical maximum frame rate and throughput for the various Ethernet topologies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igure 18.5a Theoretical throughput rates for Ethernet network topologies

7 630 TESTING NETWORK PERFORMANCE 6 F C O 6 D A H A J E? = = N E K B H = A H = J A J D H K C D F / EC = > EJ- JD A H A J. H= A H= JA M EJD K JBH= A > K HI JE C * 7 JE E = JE M EJD > K HI JE C K. = N 5. * J H = I EI I E > M E@ J D. BH= A I E A E > EJI 5 I JJE A 2 A C J D B F H A = > A 1 E JA H BH= A C = F K > A H B? I A? K JEL A BH= A I JH= I E > K HI J= BJA HBEHI JBH= A = N E K B H = A H = J A M E J D K J > K H I J E C > EJ I " ' $! ' " $ ' $ $ " B H = A I I = N E K J D H K C D F K J K I E C E E K B H = A I E > K H I J E C! " ' $ BH= A I I #! > EJI > EJ I K JE E = JE 7 JE E = JE M EJD > K HI JE K HE C > K HI J K I E C E E K B H = A I E A $ # #! $ " ' $ # $ " ' $ ' K > A H BBH= A K HE C > K HI JK I E C E E K B K '! # " ' $ % ' $ %@ K H E C ' B H = A > K = N E K B H = A H = J A M E J D > K H I J E K H E C ' B H > EJ I % " # # $ B H = A I I / EC = > EJ - J D A H A= JN E K BH= A H= JA F A H@ EHA? JE M EJD E E K. K, K F A N > EJ I # " & & ' $ ' # $ B H " = A I I F E J J F E J = N E K J D H K C D F K J F A E H A? J E M E J D E E " & & ' # BH= A I I N% # $ ' > EJI > E J I. K F A N E J D H K C D F K J M E J D E E K B H = #! & > E J I 9 EJ D = B H = A I E A B # & > O J A I J D A = N E K B A =? EHA? JE EI & % " M D E? D A G K = I ' & % K JD H K C D F K J B' & $ ' > EJ I 6 D A = N E K K J D H K C D F K J M EJ D = N E K B H = A I E A A G K = I ' % Figure 18.5b Theoretical throughput rates for Ethernet network topologies

8 Index p1 TESTING NETWORK PERFORMANCE Index of chapter 18 L Long response times 625 R Response time measurements 625 S Slow network 625 Slow server 625 T Throughput measurements 628 Throughput rates for Ethernet 629 TRACERT 627 Transmission delays 626