Scroll to navigation

IPERF(1) User Manuals IPERF(1)

NAME

iperf - perform network traffic tests using network sockets. Metrics include throughput and latency.

SYNOPSIS

iperf -s [options]

iperf -c server [options]

iperf -u -s [options]

iperf -u -c server [options]

DESCRIPTION

iperf 2 is a testing tool which performs network traffic measurements using network sockets. The performance metrics supported include throughput and latency. Iperf can use both TCP and UDP sockets (or protocols.) It supports unidirectional, full duplex (same socket) and bidirectional traffic, and supports multiple, simultaneous traffic streams. It supports multicast traffic including source specific multicast (SSM) joins. Its multi-threaded design allows for peak performance. Metrics displayed help to characterize host to host network performance. Note: Setting the enhanced (-e) option provides all available metrics.

The user must establish both a both a server (to receive traffic) and a client (to generate and send traffic) for a test to occur. The client and server typically are on different hosts or computers but need not be.

GENERAL OPTIONS

set the target bandwidth and optional standard deviation per <mean>,[<stdev>] (See NOTES for suffixes)
Display enhanced output in reports otherwise use legacy report (ver 2.0.5) formatting (see NOTES)
format to report: adaptive, bits, Bytes, Kbits, Mbits, Gbits, KBytes, MBytes, GBytes (see NOTES for more)
print a help synopsis
obscure ip addresses in output (useful when wanting to publish results and not display the full ip addresses. v4 only)
sample or display interval reports every t seconds (default) or every frame or burst, i.e. if f is used then the interval will be each frame or burst. The frame interval reporting is experimental. Also suggest a compile with fast-sampling, i.e. ./configure --enable-fastsampling
set read/write buffer size (TCP) or length (UDP) to n (TCP default 128K, UDP default 1470)
perform layer 2 length checks on received UDP packets (requires systems that support packet sockets, e.g. Linux)
print TCP maximum segment size
Override the default shared memory size between the traffic thread(s) and reporter thread in order to mitigate mutex lock contentions. The default value of 5000 should be sufficient for 1Gb/s networks. Increase this upon seeing the Warning message of reporter thread too slow. If the Warning message isn't seen, then increasing this won't have any significant effect (other than to use some additional memory.)
output the report or error message to this specified file
Set a key value that must match for the server to accept traffic on a connection. If the option is given without a value on the server a key value will be autogenerated and displayed in its initial settings report. The lifetime of the key is set using --permit-key-timeout and defaults to twenty seconds. The value is required on clients. The value will also be used as part of the transfer id in reports. The option set on the client but not the server will also cause the server to reject the client's traffic. TCP only, no UDP support.
set client or server port(s) to send or listen on per m (default 5001) w/optional port range per m-n (e.g. -p 6002-6008) (see NOTES)
sum traffic threads based upon the destination IP address (default is source ip address)
set the output to sum reports only. Useful for -P at large values
time in seconds to listen for new traffic connections, receive traffic or send traffic
use UDP rather than TCP
TCP window size (socket buffer size)
Request real-time scheduler, if supported.
bind to host, ip address or multicast address, optional port or device (see NOTES)
for use with older versions does not sent extra msgs
set TCP maximum segment size using TCP_MAXSEG
set TCP no delay, disabling Nagle's Algorithm
print version information and quit
exclude C(connection) D(data) M(multicast) S(settings) V(server) reports
if set to C or c report results as CSV (comma separated values)
Set the default congestion-control algorithm to be used for new connections. Platforms must support setsockopt's TCP_CONGESTION. (Notes: See sysctl and tcp_allowed_congestion_control for available options. May require root privileges.)

SERVER SPECIFIC OPTIONS

-1, --singleclient
set the server to process only one client at a time
set target read rate to n bits/sec. TCP only for the server.
run in server mode
enable latency histograms for udp packets (-u), for tcp writes (with --trip-times), or for either udp or tcp with --isochronous clients. The binning can be modified. Bin widths (default 1 millisecond, append u for microseconds, m for milliseconds) bincount is total bins (default 1000), ci is confidence interval between 0-100% (default lower 5%, upper 95%, 3 stdev 99.7%)
Set a key value that must match for the server to accept traffic from a client (also set with --permit-key.) The server will autogenerate a globally unique key when the option is given without a value. This value will be displayed in the server's initial settings report. The lifetime of the key is set using --permit-key-timeout and defaults to twenty seconds. TCP only, no UDP support.
Set the lifetime of the permit key in seconds. Defaults to 20 seconds if not set. A value of zero will disable the timer.
Set the receive interface to the TAP device as specified.
Set the socket option of TCP_WINDOW_CLAMP, units is bytes.
time in seconds to listen for new traffic connections and/or receive traffic (defaults to infinite)
set the socket's IP_TOS (byte) field for reverse or full duplex traffic. Supported in versions 2.1.5 or greater. Previous versions won't set IP_TOS on reverse traffic.
bind src ip addr and optional src device for receiving
run the server as a daemon. On Windows this will run the specified command-line under the IPerfService, installing the service if necessary. Note the service is not configured to auto-start or restart - if you need a self-starting service you will need to create an init script or use Windows "sc" commands.
Set the source host (ip addr) per SSM multicast, i.e. the S of the S,G
remove the IPerfService (Windows only).
run in single threaded UDP mode
Enable IPv6 reception by setting the domain and socket to AF_INET6 (Can receive on both IPv4 and IPv6)

CLIENT SPECIFIC OPTIONS

set target bandwidth to n bits/sec (default 1 Mbit/sec) or n packets per sec. This may be used with TCP or UDP. Optionally, for variable loads, use format of mean,standard deviation
run a tcp bounceback test (set size with -l or --len, defaults to 100 bytes)
request a concurrent full-duplex TCP stream
request the server to insert a delay of n milliseconds between its read and write (default is no delay)
request the client schedule a send every n milliseconds (w/o option bouncebacks are immediately scheduled. With command option and no optional value given, the interval will be one second.)
request the server not set the TCP_QUICKACK socket option (disabling TCP ACK delays) during a bounceback test (see NOTES)
Set the burst period in seconds. Defaults to one second. (Note: assumed use case is low duty cycle traffic bursts)
Set the burst size in bytes. Defaults to 1M if no value is given. -c, --client host | host%device run in client mode, connecting to host where the optional %dev will SO_BINDTODEVICE that output interface (requires root and see NOTES)
only perform a TCP connect (or 3WHS) without any data transfer, useful to measure TCP connect() times. Optional value of n is the total number of connects to do (zero is run forever.) Note that -i will rate limit the connects where -P will create bursts and -t will end the client and hence end its connect attempts.
number of times to retry a TCP connect at the application level. See operating system information on the details of TCP connect related settings.
Do a bidirectional test simultaneous test using two unidirectional sockets
Set a rate to be used with fair-queueing based socket-level pacing, in bytes or bits per second. Only available on platforms supporting the SO_MAX_PACING_RATE socket option. (Note: Here the suffixes indicate bytes/sec or bits/sec per use of uppercase or lowercase, respectively)
run a full duplex test, i.e. traffic in both transmit and receive directions using the same socket
enable select()/write() histograms with --tcp-write-prefetch. The binning can be modified. Bin widths (default 100 microseconds, append u for microseconds, m for milliseconds) bincount is total bins (default 10000), ci is confidence interval between 0-100% (default lower 5%, upper 95%, 3 stdev 99.7%)
increment the destination ip address when using the parallel (-P) or port range option
increment the destination port when using the parallel (-P) or port range option
increment the source ip address when using the parallel (-P) or port range option
increment the source ip address when using the parallel (-P) or port range option, requires -B to set the src port
set the inter-packet gap to n (units of seconds) for packets or within a frame/burst when --isochronous is set
send isochronous traffic with frequency frames per second and load defined by mean and standard deviation using a log normal distribution, defaults to 60:20m,0. (Note: Here the suffixes indicate bytes/sec or bits/sec per use of uppercase or lowercase, respectively. Also the p suffix is supported to set the burst size in packets, e.g. isochronous=2:25p will send two 25 packet bursts every second, or one 25 packet burst every 0.5 seconds.)
Set 1 to limit traffic to the local network only (through the use of SO_DONTROUTE) set to zero otherwise with optional override of compile time default (see configure --default-localonly)
Enable TCP write rate limiting per the sampled RTT. The delay is applied after the -l number of bytes have completed. The optional value is the multiplier to the RTT and defines the time delay. This value defaults to 0.5 if it is not set. Values less than 1 are supported but the value cannot be negative. This is an experimental feature. It is not likely stable on live networks. Suggested use is over controlled test networks.
By default, parallel traffic threads (per -P greater than 1) will synchronize after their TCP connects and prior to each sending traffic, i.e. all the threads first complete (or error) the TCP 3WHS before any traffic thread will start sending. This option disables that synchronization such that each traffic thread will start sending immediately after completing its successful connect.
Don't perform the UDP final server to client exchange which means there won't be a final server report displayed on the client. All packets per the test will be from the client to the server and no packets should be sent in the other direction. It's highly suggested that -t be set on the server if this option is being used. This is because there will be only one trigger ending packet sent from client to server and if it's lost then the server will continue to run. (Requires ver 2.0.14 or better)
number of bytes to transmit (instead of -t)
Set a key value that must match the server's value (also set with --permit-key) in order for the server to accept traffic from the client. TCP only, no UDP support.
Do a bidirectional test individually - client-to-server, followed by a reversed test, server-to-client
Set TCP_QUICKACK on the socket
Set TCP_NOTSENT_LOWAT on the socket and use event based writes per select() on the socket.
Measure the socket write times
time in seconds to transmit traffic, use zero for infinite (default is 10 secs)
enable the measurement of end to end write to read latencies (client and server clocks must be synchronized)
time in seconds to hold back or delay after the TCP connect and prior to the socket writes. For UDP it's the delay between the traffic thread starting and the first write.
set the txstart-time to n.n using unix or epoch time format (supports microsecond resolution, e.g 1536014418.123456) An example to delay one second using command substitution is iperf -c 192.168.1.10 --txstart-time $(expr $(date +%s) + 1).$(date +%N)
bind src ip addr and optional port as the source of traffic (see NOTES)
input the data to be transmitted from a file
input the data to be transmitted from stdin
port to receive bidirectional tests back on
number of parallel client threads to run
reverse the traffic flow (useful for testing through firewalls, see NOTES)
set the socket's IP_TOS (byte) field. Versions 2.1.5 or greater will reflect this tos setting back with --reverse or --full-duplex option. (Previous versions won't set tos on the reverse traffic.) Note: use server side --tos-override to override.
time-to-live, for multicast (default 1) -V, --ipv6_domain Set the domain to IPv6 (send packets over IPv6)
run peer version detection prior to traffic.
set TCP congestion control algorithm (Linux only)

EXAMPLES

TCP tests (client)

iperf -c <host> -e -i 1
------------------------------------------------------------
Client connecting to <host>, TCP port 5001 with pid 5149
Write buffer size: 128 KByte
TCP window size: 340 KByte (default)
------------------------------------------------------------
[ 3] local 45.56.85.133 port 49960 connected with 45.33.58.123 port 5001 (ct=3.23 ms)
[ ID] Interval Transfer Bandwidth Write/Err Rtry Cwnd/RTT NetPwr
[ 3] 0.00-1.00 sec 126 MBytes 1.05 Gbits/sec 1006/0 0 56K/626 us 210636.47
[ 3] 1.00-2.00 sec 138 MBytes 1.15 Gbits/sec 1100/0 299 483K/3884 us 37121.32
[ 3] 2.00-3.00 sec 137 MBytes 1.15 Gbits/sec 1093/0 24 657K/5087 us 28162.31
[ 3] 3.00-4.00 sec 126 MBytes 1.06 Gbits/sec 1010/0 284 294K/2528 us 52366.58
[ 3] 4.00-5.00 sec 117 MBytes 980 Mbits/sec 935/0 373 487K/2025 us 60519.66
[ 3] 5.00-6.00 sec 144 MBytes 1.20 Gbits/sec 1149/0 2 644K/3570 us 42185.36
[ 3] 6.00-7.00 sec 126 MBytes 1.06 Gbits/sec 1011/0 112 582K/5281 us 25092.56
[ 3] 7.00-8.00 sec 110 MBytes 922 Mbits/sec 879/0 56 279K/1957 us 58871.89
[ 3] 8.00-9.00 sec 127 MBytes 1.06 Gbits/sec 1014/0 46 483K/3372 us 39414.89
[ 3] 9.00-10.00 sec 132 MBytes 1.11 Gbits/sec 1054/0 0 654K/3380 us 40872.75
[ 3] 0.00-10.00 sec 1.25 GBytes 1.07 Gbits/sec 10251/0 1196 -1K/3170 us 42382.03

ct= TCP connect time (or three way handshake time 3WHS)
Write/Err Total number of successful socket writes. Total number of non-fatal socket write errors
Rtry Total number of TCP retries
Cwnd/RTT (*nix only) TCP congestion window and round trip time (sampled where NA indicates no value)
NetPwr (*nix only) Network power defined as (throughput / RTT)

TCP tests (server)

iperf -s -e -i 1 -l 8K
------------------------------------------------------------
Server listening on TCP port 5001 with pid 13430
Read buffer size: 8.00 KByte
TCP window size: 85.3 KByte (default)
------------------------------------------------------------
[ 4] local 45.33.58.123 port 5001 connected with 45.56.85.133 port 49960
[ ID] Interval Transfer Bandwidth Reads Dist(bin=1.0K)
[ 4] 0.00-1.00 sec 124 MBytes 1.04 Gbits/sec 22249 798:2637:2061:767:2165:1563:589:11669
[ 4] 1.00-2.00 sec 136 MBytes 1.14 Gbits/sec 24780 946:3227:2227:790:2427:1888:641:12634
[ 4] 2.00-3.00 sec 137 MBytes 1.15 Gbits/sec 24484 1047:2686:2218:810:2195:1819:728:12981
[ 4] 3.00-4.00 sec 126 MBytes 1.06 Gbits/sec 20812 863:1353:1546:614:1712:1298:547:12879
[ 4] 4.00-5.00 sec 117 MBytes 984 Mbits/sec 20266 769:1886:1828:589:1866:1350:476:11502
[ 4] 5.00-6.00 sec 143 MBytes 1.20 Gbits/sec 24603 1066:1925:2139:822:2237:1827:744:13843
[ 4] 6.00-7.00 sec 126 MBytes 1.06 Gbits/sec 22635 834:2464:2249:724:2269:1646:608:11841
[ 4] 7.00-8.00 sec 110 MBytes 921 Mbits/sec 21107 842:2437:2747:592:2871:1903:496:9219
[ 4] 8.00-9.00 sec 126 MBytes 1.06 Gbits/sec 22804 1038:1784:2639:656:2738:1927:573:11449
[ 4] 9.00-10.00 sec 133 MBytes 1.11 Gbits/sec 23091 1088:1654:2105:710:2333:1928:723:12550
[ 4] 0.00-10.02 sec 1.25 GBytes 1.07 Gbits/sec 227306 9316:22088:21792:7096:22893:17193:6138:120790

Reads Total number of socket reads
Dist(bin=size) Eight bin histogram of the socket reads returned byte count. Bin width is set per size. Bins are separated by a colon. In the example, the bins are 0-1K, 1K-2K, .., 7K-8K.

TCP tests (server with --trip-times on client) iperf -s -i 1 -w 4M
------------------------------------------------------------
Server listening on TCP port 5001
TCP window size: 8.00 MByte (WARNING: requested 4.00 MByte)
------------------------------------------------------------
[ 4] local 192.168.1.4%eth0 port 5001 connected with 192.168.1.7 port 44798 (trip-times) (MSS=1448) (peer 2.0.14-alpha)
[ ID] Interval Transfer Bandwidth Burst Latency avg/min/max/stdev (cnt/size) inP NetPwr Reads=Dist
[ 4] 0.00-1.00 sec 19.0 MBytes 159 Mbits/sec 52.314/10.238/117.155/19.779 ms (151/131717) 1.05 MByte 380.19 781=306:253:129:48:18:15:8:4
[ 4] 1.00-2.00 sec 20.0 MBytes 168 Mbits/sec 53.863/21.264/79.252/12.277 ms (160/131080) 1.08 MByte 389.38 771=294:236:126:60:18:24:10:3
[ 4] 2.00-3.00 sec 18.2 MBytes 153 Mbits/sec 58.718/22.000/137.944/20.397 ms (146/130964) 1.06 MByte 325.64 732=299:231:98:52:18:19:10:5
[ 4] 3.00-4.00 sec 19.7 MBytes 165 Mbits/sec 50.448/ 8.921/82.728/14.627 ms (158/130588) 997 KByte 409.00 780=300:255:121:58:15:18:7:6
[ 4] 4.00-5.00 sec 18.8 MBytes 158 Mbits/sec 53.826/11.169/115.316/15.541 ms (150/131420) 1.02 MByte 366.24 761=302:226:134:52:22:17:7:1
[ 4] 5.00-6.00 sec 19.5 MBytes 164 Mbits/sec 50.943/11.922/76.134/14.053 ms (156/131276) 1.03 MByte 402.00 759=273:246:149:45:16:18:4:8
[ 4] 6.00-7.00 sec 18.5 MBytes 155 Mbits/sec 57.643/10.039/127.850/18.950 ms (148/130926) 1.05 MByte 336.16 710=262:228:133:37:16:20:8:6
[ 4] 7.00-8.00 sec 19.6 MBytes 165 Mbits/sec 52.498/12.900/77.045/12.979 ms (157/131003) 1.00 MByte 391.78 742=288:200:135:68:16:23:4:8
[ 4] 8.00-9.00 sec 18.0 MBytes 151 Mbits/sec 58.370/ 8.026/150.243/21.445 ms (144/131255) 1.06 MByte 323.81 716=268:241:108:51:20:17:8:3
[ 4] 9.00-10.00 sec 18.4 MBytes 154 Mbits/sec 56.112/12.419/79.790/13.668 ms (147/131194) 1.05 MByte 343.70 822=330:303:120:26:16:14:9:4
[ 4] 10.00-10.06 sec 1.03 MBytes 146 Mbits/sec 69.880/45.175/78.754/10.823 ms (9/119632) 1.74 MByte 260.40 62=26:30:5:1:0:0:0:0
[ 4] 0.00-10.06 sec 191 MBytes 159 Mbits/sec 54.183/ 8.026/150.243/16.781 ms (1526/131072) 1.03 MByte 366.98 7636=2948:2449:1258:498:175:185:75:48

Burst Latency One way TCP write() to read() latency in mean/minimum/maximum/standard deviation format (Note: requires the client's and server's system clocks to be synchronized to a common reference, e.g. using precision time protocol PTP. A GPS disciplined OCXO is a recommended reference.)
cnt Number of completed bursts received and used for the burst latency calculations
size Average burst size in bytes (computed average and estimate only)
inP inP, short for in progress, is the average number of bytes in progress or in flight. This is taken from the application level write to read perspective. Note this is a mean value. The parenthesis value is the standard deviation from the mean. (Requires --trip-times on client. See Little's law in NOTES.)
NetPwr Network power defined as (throughput / one way latency)

TCP tests (with one way delay sync check -X and --trip-times on the client)

iperf -c 192.168.1.4 -X -e --trip-times -i 1 -t 2
------------------------------------------------------------
Client connecting to 192.168.1.4, TCP port 5001 with pid 16762 (1 flows)
Write buffer size: 131072 Byte
TCP window size: 85.0 KByte (default)
------------------------------------------------------------
[ 1] Clock sync check (ms): RTT/Half=(3.361/1.680) OWD-send/ack/asym=(2.246/1.115/1.131)
[ 1] local 192.168.1.1%ap0 port 47466 connected with 192.168.1.4 port 5001 (MSS=1448) (trip-times) (sock=3) (peer 2.1.4-master)
[ ID] Interval Transfer Bandwidth Write/Err Rtry Cwnd/RTT NetPwr
[ 1] 0.00-1.00 sec 9.50 MBytes 79.7 Mbits/sec 77/0 0 2309K/113914 us 87
[ 1] 1.00-2.00 sec 7.12 MBytes 59.8 Mbits/sec 57/0 0 2492K/126113 us 59
[ 1] 2.00-2.42 sec 128 KBytes 2.47 Mbits/sec 2/0 0 2492K/126113 us 2
[ 1] 0.00-2.42 sec 16.8 MBytes 58.0 Mbits/sec 136/0 0 2492K/126113 us 57

UDP tests (client)

iperf -c <host> -e -i 1 -u -b 10m
------------------------------------------------------------
Client connecting to <host>, UDP port 5001 with pid 5169
Sending 1470 byte datagrams, IPG target: 1176.00 us (kalman adjust)
UDP buffer size: 208 KByte (default)
------------------------------------------------------------
[ 3] local 45.56.85.133 port 32943 connected with 45.33.58.123 port 5001
[ ID] Interval Transfer Bandwidth Write/Err PPS
[ 3] 0.00-1.00 sec 1.19 MBytes 10.0 Mbits/sec 852/0 851 pps
[ 3] 1.00-2.00 sec 1.19 MBytes 10.0 Mbits/sec 850/0 850 pps
[ 3] 2.00-3.00 sec 1.19 MBytes 10.0 Mbits/sec 850/0 850 pps
[ 3] 3.00-4.00 sec 1.19 MBytes 10.0 Mbits/sec 851/0 850 pps
[ 3] 4.00-5.00 sec 1.19 MBytes 10.0 Mbits/sec 850/0 850 pps
[ 3] 5.00-6.00 sec 1.19 MBytes 10.0 Mbits/sec 850/0 850 pps
[ 3] 6.00-7.00 sec 1.19 MBytes 10.0 Mbits/sec 851/0 850 pps
[ 3] 7.00-8.00 sec 1.19 MBytes 10.0 Mbits/sec 850/0 850 pps
[ 3] 8.00-9.00 sec 1.19 MBytes 10.0 Mbits/sec 851/0 850 pps
[ 3] 0.00-10.00 sec 11.9 MBytes 10.0 Mbits/sec 8504/0 850 pps
[ 3] Sent 8504 datagrams
[ 3] Server Report:
[ 3] 0.00-10.00 sec 11.9 MBytes 10.0 Mbits/sec 0.047 ms 0/ 8504 (0%) 0.537/ 0.392/23.657/ 0.497 ms 850 pps 2329.37

Write/Err Total number of successful socket writes. Total number of non-fatal socket write errors
PPS Transmit packet rate in packets per second

UDP tests (server) iperf -s -i 1 -w 4M -u
------------------------------------------------------------
Server listening on UDP port 5001
Receiving 1470 byte datagrams
UDP buffer size: 8.00 MByte (WARNING: requested 4.00 MByte)
------------------------------------------------------------
[ 3] local 192.168.1.4 port 5001 connected with 192.168.1.1 port 60027 (WARN: winsize=8.00 MByte req=4.00 MByte) (trip-times) (0.0) (peer 2.0.14-alpha)
[ ID] Interval Transfer Bandwidth Jitter Lost/Total Latency avg/min/max/stdev PPS inP NetPwr
[ 3] 0.00-1.00 sec 44.5 MBytes 373 Mbits/sec 0.071 ms 52198/83938 (62%) 75.185/ 2.367/85.189/14.430 ms 31854 pps 3.64 MByte 620.58
[ 3] 1.00-2.00 sec 44.8 MBytes 376 Mbits/sec 0.015 ms 59549/143701 (41%) 79.609/75.603/85.757/ 1.454 ms 31954 pps 3.56 MByte 590.04
[ 3] 2.00-3.00 sec 44.5 MBytes 373 Mbits/sec 0.017 ms 59494/202975 (29%) 80.006/75.951/88.198/ 1.638 ms 31733 pps 3.56 MByte 583.07
[ 3] 3.00-4.00 sec 44.5 MBytes 373 Mbits/sec 0.019 ms 59586/262562 (23%) 79.939/75.667/83.857/ 1.145 ms 31767 pps 3.56 MByte 583.57
[ 3] 4.00-5.00 sec 44.5 MBytes 373 Mbits/sec 0.081 ms 59612/322196 (19%) 79.882/75.400/86.618/ 1.666 ms 31755 pps 3.55 MByte 584.40
[ 3] 5.00-6.00 sec 44.7 MBytes 375 Mbits/sec 0.064 ms 59571/381918 (16%) 79.767/75.571/85.339/ 1.556 ms 31879 pps 3.56 MByte 588.02
[ 3] 6.00-7.00 sec 44.6 MBytes 374 Mbits/sec 0.041 ms 58990/440820 (13%) 79.722/75.662/85.938/ 1.087 ms 31820 pps 3.58 MByte 586.73
[ 3] 7.00-8.00 sec 44.7 MBytes 375 Mbits/sec 0.027 ms 59679/500548 (12%) 79.745/75.704/84.731/ 1.094 ms 31869 pps 3.55 MByte 587.46
[ 3] 8.00-9.00 sec 44.3 MBytes 371 Mbits/sec 0.078 ms 59230/559499 (11%) 80.346/75.514/94.293/ 2.858 ms 31590 pps 3.58 MByte 577.97
[ 3] 9.00-10.00 sec 44.4 MBytes 373 Mbits/sec 0.073 ms 58782/618394 (9.5%) 79.125/75.511/93.638/ 1.643 ms 31702 pps 3.55 MByte 588.99
[ 3] 10.00-10.08 sec 3.53 MBytes 367 Mbits/sec 0.129 ms 6026/595236 (1%) 94.967/80.709/99.685/ 3.560 ms 31107 pps 3.58 MByte 483.12
[ 3] 0.00-10.08 sec 449 MBytes 374 Mbits/sec 0.129 ms 592717/913046 (65%) 79.453/ 2.367/99.685/ 5.200 ms 31776 pps (null) 587.91

Latency End to end latency in mean/minimum/maximum/standard deviation format (Note: requires the client's and server's system clocks to be synchronized to a common reference, e.g. using precision time protocol PTP. A GPS disciplined OCXO is a recommended reference.)
PPS Received packet rate in packets per second
inP inP, short for in progress, is the average number of bytes in progress or in flight. This is taken from an application write to read perspective. (Requires --trip-times on client. See Little's law in NOTES.)
NetPwr Network power defined as (throughput / latency)

Isochronous UDP tests (client)

iperf -c 192.168.100.33 -u -e -i 1 --isochronous=60:100m,10m --realtime
------------------------------------------------------------
Client connecting to 192.168.100.33, UDP port 5001 with pid 14971
UDP isochronous: 60 frames/sec mean= 100 Mbit/s, stddev=10.0 Mbit/s, Period/IPG=16.67/0.005 ms
UDP buffer size: 208 KByte (default)
------------------------------------------------------------
[ 3] local 192.168.100.76 port 42928 connected with 192.168.100.33 port 5001
[ ID] Interval Transfer Bandwidth Write/Err PPS frames:tx/missed/slips
[ 3] 0.00-1.00 sec 12.0 MBytes 101 Mbits/sec 8615/0 8493 pps 62/0/0
[ 3] 1.00-2.00 sec 12.0 MBytes 100 Mbits/sec 8556/0 8557 pps 60/0/0
[ 3] 2.00-3.00 sec 12.0 MBytes 101 Mbits/sec 8586/0 8586 pps 60/0/0
[ 3] 3.00-4.00 sec 12.1 MBytes 102 Mbits/sec 8687/0 8687 pps 60/0/0
[ 3] 4.00-5.00 sec 11.8 MBytes 99.2 Mbits/sec 8468/0 8468 pps 60/0/0
[ 3] 5.00-6.00 sec 11.9 MBytes 99.8 Mbits/sec 8519/0 8520 pps 60/0/0
[ 3] 6.00-7.00 sec 12.1 MBytes 102 Mbits/sec 8694/0 8694 pps 60/0/0
[ 3] 7.00-8.00 sec 12.1 MBytes 102 Mbits/sec 8692/0 8692 pps 60/0/0
[ 3] 8.00-9.00 sec 11.9 MBytes 100 Mbits/sec 8537/0 8537 pps 60/0/0
[ 3] 9.00-10.00 sec 11.8 MBytes 99.0 Mbits/sec 8450/0 8450 pps 60/0/0
[ 3] 0.00-10.01 sec 120 MBytes 100 Mbits/sec 85867/0 8574 pps 602/0/0
[ 3] Sent 85867 datagrams
[ 3] Server Report:
[ 3] 0.00-9.98 sec 120 MBytes 101 Mbits/sec 0.009 ms 196/85867 (0.23%) 0.665/ 0.083/ 1.318/ 0.174 ms 8605 pps 18903.85

frames:tx/missed/slips Total number of isochronous frames or bursts. Total number of frame ids not sent. Total number of frame slips

Isochronous UDP tests (server)

iperf -s -e -u --udp-histogram=100u,2000 --realtime
------------------------------------------------------------
Server listening on UDP port 5001 with pid 5175
Receiving 1470 byte datagrams
UDP buffer size: 208 KByte (default)
------------------------------------------------------------
[ 3] local 192.168.100.33 port 5001 connected with 192.168.100.76 port 42928 isoch (peer 2.0.13-alpha)
[ ID] Interval Transfer Bandwidth Jitter Lost/Total Latency avg/min/max/stdev PPS NetPwr Frames/Lost
[ 3] 0.00-9.98 sec 120 MBytes 101 Mbits/sec 0.010 ms 196/85867 (0.23%) 0.665/ 0.083/ 1.318/ 0.284 ms 8585 pps 18903.85 601/1
[ 3] 0.00-9.98 sec T8(f)-PDF: bin(w=100us):cnt(85671)= 1:2,2:844,3:10034,4:8493,5:8967,6:8733,7:8823,8:9023,9:8901,10:8816,11:7730,12:4563,13:741,14:1 (5.00/95.00%=3/12,Outliers=0,obl/obu=0/0)
[ 3] 0.00-9.98 sec F8(f)-PDF: bin(w=100us):cnt(598)= 15:2,16:1,17:27,18:68,19:125,20:136,21:103,22:83,23:22,24:23,25:5,26:3 (5.00/95.00%=17/24,Outliers=0,obl/obu=0/0)

Frames/lost Total number of frames (or bursts) received. Total number of bursts lost or error-ed
T8-PDF(f) Latency histogram for packets
F8-PDF(f) Latency histogram for frames

ENVIRONMENT

The environment variable option settings haven't been maintained well. See the source code if these are of interest.

NOTES

Numeric options: Some numeric options support format characters per '<value>c' (e.g. 10M) where the c format characters are k,m,g,K,M,G. Lowercase format characters are 10^3 based and uppercase are 2^n based, e.g. 1k = 1000, 1K = 1024, 1m = 1,000,000 and 1M = 1,048,576

Rate limiting: The -b option supports read and write rate limiting at the application level. The -b option on the client also supports variable offered loads through the <mean>,<standard deviation> format, e.g. -b 100m,10m. The distribution used is log normal. Similar for the isochronous option. The -b on the server rate limits the reads. Socket based pacing is also supported using the --fq-rate long option. This will work with the --reverse and --full-duplex options as well.

Synchronized clocks: The --trip-times option indicates that the client's and server's clocks are synchronized to a common reference. Network Time Protocol (NTP) or Precision Time Protocol (PTP) are commonly used for this. The reference clock(s) error and the synchronization protocols will affect the accuracy of any end to end latency measurements.

Histograms and non-parametric statisitics: The --histograms option provides the raw data where nothing is averaged. This is useful for non-parametric distributions, e.g. latency. The standard output does use the central limit theorem to produce average, mininimum, maximum and variation. This loses information when the underlining distribution is not gaussian. Histograms are supported so this information is made available.

Binding is done at the logical level of port and ip address (or layer 3) using the -B option and a colon as the separator between port and the ip addr. Binding at the device (or layer 2) level requires the percent (%) as the delimiter (for both the client and the server.) An example for src port and ip address is -B 192.168.1.1:6001. To bind the src port only and let the operating system choose the source ip address use 0.0.0.0, e.g. -B 0.0.0.0:6001. On the client, the -B option affects the bind(2) system call, and will set the source ip address and the source port, e.g. iperf -c <host> -B 192.168.100.2:6002. This controls the packet's source values but not routing. These can be confusing in that a route or device lookup may not be that of the device with the configured source IP. So, for example, if the IP address of eth0 is used for -B and the routing table for the destination IP address resolves the output interface to be eth1, then the host will send the packet out device eth1 while using the source IP address of eth0 in the packet. To affect the physical output interface (e.g. dual homed systems) either use -c <host>%<dev> (requires root) which bypasses this host route table lookup, or configure policy routing per each -B source address and set the output interface appropriately in the policy routes. On the server or receive, only packets destined to -B IP address will be received. It's also useful for multicast. For example, iperf -s -B 224.0.0.1%eth0 will only accept ip multicast packets with dest ip 224.0.0.1 that are received on the eth0 interface, while iperf -s -B 224.0.0.1 will receive those packets on any interface, Finally, the device specifier is required for v6 link-local, e.g. -c [v6addr]%<dev> -V, to select the output interface.

Reverse, full-duplex, dualtest (-d) and tradeoff (-r): The --reverse (-R) and --full-duplex options can be confusing when compared to the older options of --dualtest (-d) and --tradeoff (-r). The newer options of --reverse and --full-duplex only open one socket and read and write to the same socket descriptor, i.e. use the socket in full duplex mode. The older -d and -r open second sockets in the opposite direction and do not use a socket in full duplex mode. Note that full duplex applies to the socket and not to the network devices and that full duplex sockets are supported by the operating systems regardless if an underlying network supports full duplex transmission and reception. It's suggested to use --reverse if you want to test through a NAT firewall (or -R on non-windows systems). This applies role reversal of the test after opening the full duplex socket. (Note: Firewall piercing may be required to use -d and -r if a NAT gateway is in the path.)

Also, the --reverse -b <rate> setting behaves differently for TCP and UDP. For TCP it will rate limit the read side, i.e. the iperf client (role reversed to act as a server) reading from the full duplex socket. This will in turn flow control the reverse traffic per standard TCP congestion control. The --reverse -b <rate> will be applied on transmit (i.e. the server role reversed to act as a client) for UDP since there is no flow control with UDP. There is no option to directly rate limit the writes with TCP testing when using --reverse.

Bounceback The bounceback test allows one to measure network responsiveness (which, in this test, is an inverse of latency.) Latency is merely delay in units of time. Latency metrics require one to know the delay of what's being measured. For bounceback it's a client write to a server read followed by a server write and then the client read. The original write is bounce backed. Iperf 2 sets up the socket with TCP_NODELAY and possibly TCP_QUICKACK (unless disabled). The client sends a small write (which defaults to 100 bytes unless -l is set) and issues a read waiting for the "bounceback" from the server. The server waits for a read and then optionally delays before sending the payload back. This repeats until the traffic ends.

The TCP_QUICKACK socket option will be enabled during bounceback tests when the bounceback-hold is set to a non-zero value. The socket option is applied after every read() on the server and before the hold delay call. It's also applied on the client. Use --bounceback-no-quickack to disable TCP ack delays per the socket.

TCP Connect times: The TCP connect time (or three way handshake) can be seen on the iperf client when the -e (--enhanced) option is set. Look for the ct=<value> in the connected message, e.g.in '[ 3] local 192.168.1.4 port 48736 connected with 192.168.1.1 port 5001 (ct=1.84 ms)' shows the 3WHS took 1.84 milliseconds.

Port-range Port ranges are supported using the hyphen notation, e.g. 6001-6009. This will cause multiple threads, one per port, on either the listener/server or the client. The user needs to take care that the ports in the port range are available and not already in use per the operating system. The -P is supported on the client and will apply to each destination port within the port range. Finally, this can be used for a workaround for Windows UDP and -P > 1 as Windows doesn't dispatch UDP per a server's connect and the quintuple.

Packet per second (pps) calculation The packets per second calculation is done as a derivative, i.e. number of packets divided by time. The time is taken from the previous last packet to the current last packet. It is not the sample interval time. The last packet can land at different times within an interval. This means that pps does not have to match rx bytes divided by the sample interval. Also, with --trip-times set, the packet time on receive is set by the sender's write time so pps indicates the end to end pps with --trip-times. The RX pps calculation is receive side only when -e is set and --trip-times is not set.

Little's Law in queuing theory is a theorem that determines the average number of items (L) in a stationary queuing system based on the average waiting time (W) of an item within a system and the average number of items arriving at the system per unit of time (lambda). Mathematically, it's L = lambda * W. As used here, the units are bytes. The arrival rate is taken from the writes.

Network power: The network power (NetPwr) metric is experimental. It's a convenience function defined as throughput/delay. For TCP transmits, the delay is the sampled RTT times. For TCP receives, the delay is the write to read latency. For UDP the delay is the end/end latency. Don't confuse this with the physics definition of power (delta energy/delta time) but more of a measure of a desirable property divided by an undesirable property. Also note, one must use -i interval with TCP to get this as that's what sets the RTT sampling rate. The metric is scaled to assist with human readability.

Multicast: Iperf 2 supports multicast with a couple of caveats. First, multicast streams cannot take advantage of the -P option. The server will serialize multicast streams. Also, it's highly encouraged to use a -t on a server that will be used for multicast clients. That is because the single end of traffic packet sent from client to server may get lost and there are no redundant end of traffic packets. Setting -t on the server will kill the server thread in the event this packet is indeed lost.

TCP_QUICACK: The TCP_QUICKACK socket applied after every read() on the server

Fast Sampling: Use ./configure --enable-fastsampling and then compile from source to enable four digit (e.g. 1.0000) precision in reports' timestamps. Useful for sub-millisecond sampling.

DIAGNOSTICS

Use ./configure --enable-thread-debug and then compile from source to enable both asserts and advanced debugging of the tool itself.

BUGS

See https://sourceforge.net/p/iperf2/tickets/

AUTHORS

Iperf2, based from iperf (originally written by Mark Gates and Alex Warshavsky), has a goal of maintenance with some feature enhancement. Other contributions from Ajay Tirumala, Jim Ferguson, Jon Dugan <jdugan at x1024 dot net>, Feng Qin, Kevin Gibbs, John Estabrook <jestabro at ncsa.uiuc.edu>, Andrew Gallatin <gallatin at gmail.com>, Stephen Hemminger <shemminger at linux-foundation.org>, Tim Auckland <tim.auckland at gmail.com>, Robert J. McMahon <rjmcmahon at rjmcmahon.com>

SEE ALSO

accept(2),bind(2),close(2),connect(2),fcntl(2),getpeername(2),getsockname(2),getsockopt(2),listen(2),read(2),recv(2),select(2),send(2),setsockopt(2),shutdown(2),write(2),ip(7),socket(7),tcp(7),udp(7)

Source code at http://sourceforge.net/projects/iperf2/

"Unix Network Programming, Volume 1: The Sockets Networking API (3rd Edition) 3rd Edition" by W. Richard Stevens (Author), Bill Fenner (Author), Andrew M. Rudoff (Author)

January 2022 NLANR/DAST