Provided by: iperf_2.2.1+dfsg-1_amd64 
NAME
iperf - perform network traffic tests using network sockets. Metrics include throughput and latency or
link capacity and responsiveness.
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 (or link capacity and responsiveness.)
Latency measurements include both one way delay (OWD) and round trip times (RTTs.) 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. Setting the enhanced (-e) option
provides all available metrics. Note: the metrics are at the socket level reads and writes. They do not
include the overhead associated with lower level protocol layer headers.
The user must establish 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
-b, --bandwidth
set the target bandwidth and optional standard deviation per <mean>,[<stdev>] (See NOTES for
suffixes) Setting the target bitrate on the client to 0 will disable bitrate limits (particularly
useful for UDP tests). Will limit the read rate on the server.
-e, --enhanced
Display enhanced output in reports otherwise use legacy report (ver 2.0.5) formatting (see NOTES)
-f, --format [abkmgBKMG]
format to report: adaptive, bits, Bytes, Kbits, Mbits, Gbits, KBytes, MBytes, GBytes (see NOTES
for more)
-h, --help
print a help synopsis
--hide-ips
obscure ip addresses in output (useful when wanting to publish results and not display the full ip
addresses. v4 only)
-i, --interval < t | f >
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
-l, --len n[kmKM]
set read/write buffer size (TCP) or length (UDP) to n (TCP default 128K, UDP default 1470)
--l2checks
perform layer 2 length checks on received UDP packets (requires systems that support packet
sockets, e.g. Linux)
-m, --print_mss
print TCP maximum segment size
--NUM_REPORT_STRUCTS <count>
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.)
-o, --output filename
output the report or error message to this specified file
--permit-key [=<value>]
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 on clients required the use of '=', e.g. --permit-key=password (even though
it's required command line option.) The server will auto-generate a value if '=password' is not
given. 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.
-p, --port m[-n]
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-dstip
sum traffic threads based upon the destination IP address (default is source ip address)
--sum-only
set the output to sum reports only. Useful for -P at large values
--tcp-tx-delay n
Set TCP_TX_DELAY on the socket. Delay units are milliseconds. Value takes float. See Notes for
qdisc requirements.
-t, --time n
time in seconds to listen for new traffic connections, receive traffic or send traffic
-u, --udp
use UDP rather than TCP
--utc
use coordinated universal time (UTC) when outputting time (otherwise use local time)
-w, --window n[kmKM]
TCP window size (socket buffer size)
-z, --realtime
Request real-time scheduler, if supported.
-B, --bind host[:port][%dev]
bind to host, ip address or multicast address, optional port or device (see NOTES)
-C, --compatibility
for use with older versions does not sent extra msgs
-M, --mss n
set TCP maximum segment size using TCP_MAXSEG
-N, --nodelay
set TCP no delay, disabling Nagle's Algorithm
-v, --version
print version information and quit
-x, --reportexclude [CDMSV]
exclude C(connection) D(data) M(multicast) S(settings) V(server) reports
-y, --reportstyle C|c
if set to C or c report results as CSV (comma separated values)
--tcp-cca
Set the congestion control algorithm to be used for TCP connections. See SPECIFIC OPTIONS for more
--working-load-cca
Set the congestion control algorithm to be used for TCP working loads. See SPECIFIC OPTIONS for
more
-Z, --tcp-congestion
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
-b, --bandwidth n[kmgKMG]
set target read rate to n bits/sec. TCP only for the server.
-s, --server
run in server mode
--histograms[=binwidth[u],bincount,[lowerci],[upperci]]
enable latency histograms for udp packets (-u), for tcp writes (with --trip-times), or for either
udp or tcp with --isochronous clients, or for --bounceback. 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%)
--jitter-histograms[=<binwidth>]
enable jitter histograms for udp packets (-u). Optional value is the bin width where units are
microseconds and defaults to 100 usecs
--permit-key [=<value>]
Set a key value that must match for the server to accept traffic from a client (also set with
--permit-key.) The server will auto-generate 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.
--permit-key-timeout <value>
Set the lifetime of the permit key in seconds. Defaults to 20 seconds if not set. A value of zero
will disable the timer.
--skip-rx-copy
Set the server threads to use MSG_TRUNC on recv when possible. This flag causes the received bytes
of data to be discarded and should offload the receiving CPU some.
--tap-dev <dev>
Set the receive interface to the TAP device as specified.
--tcp-rx-window-clamp n[kmKM]
Set the socket option of TCP_WINDOW_CLAMP, units is bytes.
--test-exchange-timeout <value>
Set the maximum wait time for a test exchange in seconds. Defaults to 60 seconds if not set. A
value of zero will disable the timeout.
-t, --time n
time in seconds to listen for new traffic connections and/or receive traffic (defaults to
infinite)
--tos-override <val>
set the socket's IP_TOS value for reverse or full duplex traffic. Supported in versions 2.1.5 or
greater. Previous versions won't set IP_TOS on reverse traffic. See NOTES for values.
-B, --bind ip | ip%device
bind src ip addr and optional src device for receiving
-D, --daemon
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.
-H, --ssm-host host
Set the source host (ip addr) per SSM multicast, i.e. the S of the S,G
-R, --remove
remove the IPerfService (Windows only).
-U, --single_udp
run in single threaded UDP mode
-V, --ipv6_domain
Enable IPv6 reception by setting the domain and socket to AF_INET6 (Can receive on both IPv4 and
IPv6)
--tcp-cca
Set the congestion control algorithm to be used for TCP connections - will override any client
side settings (same as --tcp-congestion)
--working-load
Enable support for TCP working loads on UDP traffic streams
--working-load-cca
Set the congestion control algorithm to be used for TCP working loads - will override any client
side settings
CLIENT SPECIFIC OPTIONS
-b, --bandwidth n[kmgKMG][,n[kmgKMG]] | n[kmgKMG]pps
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
--bounceback[=n]
run a TCP bounceback or rps test with optional number writes in a burst per value of n. The
default is ten writes every period and the default period is one second (Note: set size with
--bounceback-request). See NOTES on clock unsynchronized detections.
--bounceback-hold n
request the server to insert a delay of n milliseconds between its read and write (default is no
delay)
--bounceback-no-quickack
request the server not set the TCP_QUICKACK socket option (disabling TCP ACK delays) during a
bounceback test (see NOTES)
--bounceback-period[=n]
request the client schedule its send(s) every n seconds (default is one second, use zero value for
immediate or continuous back to back)
--bounceback-request n
set the bounceback request size in units bytes. Default value is 100 bytes.
--bounceback-reply n
set the bounceback reply size in units bytes. This supports asymmetric message sizes between the
request and the reply. Default value is zero, which uses the value of --bounceback-request.
--bounceback-txdelay n
request the client to delay n seconds between the start of the working load and the bounceback
traffic (default is no delay)
--burst-period n
Set the burst period in seconds. Defaults to one second. (Note: assumed use case is low duty cycle
traffic bursts)
--burst-size n
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)
--connect-only[=n]
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.
--connect-retry-time n
time value in seconds for application level retries of TCP connect(s). See --connect-retry-timer
for the retry time interval. See operating system information for the details of system or kernel
TCP connect related settings. This is an application level retry of the connect() call and not the
system level connect.
--connect-retry-timer n
The minimum time value in seconds to wait before retrying the connect. Note: This a minimum time
to wait between retries and can be longer dependent upon the system connect time taken. See
operating system information for the details of system or kernel TCP connect related settings.
--dscp
set the DSCP field (masking ECN bits) in the TOS byte (used by IP_TOS & setsockopt)
-d, --dualtest
Do a bidirectional test simultaneous test using two unidirectional sockets
--fq-rate n[kmgKMG]
Set a rate to be used with fair-queuing 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)
--fq-rate-step n[kmgKMG]
Set a step of rate to be used with fair-queuing based socket-level pacing, in bytes or bits per
second. Step occurs every fq-rate-step-interval (defaults to one second)
--fq-rate-step-interval n
Time in seconds before stepping the fq-rate
--full-duplex
run a full duplex test, i.e. traffic in both transmit and receive directions using the same socket
--histograms[=binwidth[u],bincount,[lowerci],[upperci]]
enable select()/write() histograms with --tcp-write-times or --bounceback (these options are
mutually exclusive.) 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%)
--ignore-shutdown
don't wait on the TCP shutdown or close (fin & finack) rather use the final write as the ending
event
--incr-dstip
increment the destination ip address when using the parallel (-P) or port range option
--incr-dstport
increment the destination port when using the parallel (-P) or port range option
--incr-srcip
increment the source ip address when using the parallel (-P) or port range option
--incr-srcport
increment the source ip address when using the parallel (-P) or port range option, requires -B to
set the src port
--ipg n
set the inter-packet gap to n (units of seconds) for packets or within a frame/burst when
--isochronous is set
--isochronous[=fps:mean,stdev]
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.)
--local-only[=1|0]
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)
--near-congestion[=n]
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.
--no-connect-sync
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.
--no-udp-fin
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)
-n, --num n[kmKM]
number of bytes to transmit (instead of -t)
--permit-key [=<value>]
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.
--sync-transfer-id
Pass the clients' transfer id(s) to the server so both will use the same id in their respective
outputs
-r, --tradeoff
Do a bidirectional test individually - client-to-server, followed by a reversed test,
server-to-client
--tcp-cca
Set the congestion control algorithm to be used for TCP connections & exchange with the server
(same as --tcp-congestion)
--tcp-quickack
Set TCP_QUICKACK on the socket
--tcp-write-prefetch n[kmKM]
Set TCP_NOTSENT_LOWAT on the socket and use event based writes per select() on the socket.
--tcp-write-times
Measure the socket write times
-t, --time n|0
time in seconds to transmit traffic, use zero for infinite (default is 10 secs)
--trip-times
enable the measurement of end to end write to read latencies (client and server clocks must be
synchronized.) See notes about tcp-write-prefetch being enabled.
--txdelay-time
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.
--txstart-time n.n
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)
--udp-l4s
run an l4s traffic load (requires a iperf server that supports l4s)
-B, --bind ip | ip:port | ipv6 -V | [ipv6]:port -V
bind src ip addr and optional port as the source of traffic (see NOTES)
-F, --fileinput name
input the data to be transmitted from a file
-I, --stdin
input the data to be transmitted from stdin
-L, --listenport n
port to receive bidirectional tests back on
-P, --parallel n
number of parallel client threads to run
-R, --reverse
reverse the traffic flow (useful for testing through firewalls, see NOTES)
-S, --tos <val>
set the socket's IP_TOS value. 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. See NOTES for values.
-T, --ttl n
time-to-live, for multicast (default 1)
--working-load[=up|down|bidir][,n]
request a concurrent working load, currently TCP stream(s), defaults to full duplex (or bidir)
unless the up or down option is provided. The number of TCP streams defaults to 1 and can be
changed via the n value, e.g. --working-load=down,4 will use four TCP streams from server to the
client as the working load. The IP ToS will be BE (0x0) for working load traffic.
--working-load-cca
Set the congestion control algorithm to be used for TCP working loads, exchange with the server
-V, --ipv6_domain
Set the domain to IPv6 (send packets over IPv6)
-X, --peerdetect
run peer version detection prior to traffic.
-Z, --linux-congestion algo
set TCP congestion control algorithm (Linux only)
EXAMPLES
TCP tests (client) iperf -c <host> -e -i 1
------------------------------------------------------------
Client connecting to 192.168.1.35, TCP port 5001 with pid 256370 (1/0 flows/load)
Write buffer size: 131072 Byte
TCP congestion control using cubic
TOS set to 0x0 (dscp=0,ecn=0) (Nagle on)
TCP window size: 100 MByte (default)
------------------------------------------------------------
[ 1] local 192.168.1.103%enp4s0 port 41024 connected with 192.168.1.35 port 5001 (sock=3)
(icwnd/mss/irtt=14/1448/158) (ct=0.21 ms) on 2024-03-26 10:48:47.867 (PDT)
[ ID] Interval Transfer Bandwidth Write/Err Rtry InF(pkts)/Cwnd(pkts)/RTT(var)
NetPwr
[ 1] 0.00-1.00 sec 201 MBytes 1.68 Gbits/sec 1605/0 73 1531K(1083)/1566K(1108)/13336(112)
us 15775
[ 1] 1.00-2.00 sec 101 MBytes 846 Mbits/sec 807/0 0 1670K(1181)/1689K(1195)/14429(83)
us 7331
[ 1] 2.00-3.00 sec 101 MBytes 847 Mbits/sec 808/0 0 1790K(1266)/1790K(1266)/15325(97)
us 6911
[ 1] 3.00-4.00 sec 134 MBytes 1.13 Gbits/sec 1075/0 0 1858K(1314)/1892K(1338)/16188(99)
us 8704
[ 1] 4.00-5.00 sec 101 MBytes 846 Mbits/sec 807/0 1 1350K(955)/1370K(969)/11620(98) us
9103
[ 1] 5.00-6.00 sec 121 MBytes 1.01 Gbits/sec 966/0 0 1422K(1006)/1453K(1028)/12405(118)
us 10207
[ 1] 6.00-7.00 sec 115 MBytes 962 Mbits/sec 917/0 0 1534K(1085)/1537K(1087)/13135(105)
us 9151
[ 1] 7.00-8.00 sec 101 MBytes 844 Mbits/sec 805/0 0 1532K(1084)/1580K(1118)/13582(136)
us 7769
[ 1] 8.00-9.00 sec 134 MBytes 1.13 Gbits/sec 1076/0 0 1603K(1134)/1619K(1145)/13858(105)
us 10177
[ 1] 9.00-10.00 sec 101 MBytes 846 Mbits/sec 807/0 0 1602K(1133)/1650K(1167)/14113(105)
us 7495
[ 1] 10.00-10.78 sec 128 KBytes 1.34 Mbits/sec 1/0 0 0K(0)/1681K(1189)/14424(111) us
11.64
[ 1] 0.00-10.78 sec 1.18 GBytes 941 Mbits/sec 9674/0 74 0K(0)/1681K(1189)/14424(111)
us 8154
where (per -e,)
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
Inf(pkts)/Cwnd(pkts)/RTT(var) (*nix only) TCP bytes and packets inflight, congestion window bytes
and packets, and round trip time and variance (sampled where NA indicates no value). Inflight is
in units of Kbytes and packets where packets_in_flight = (tcp_info_buf.tcpi_unacked -
tcp_info_buf.tcpi_sacked - tcp_info_buf.tcpi_lost + tcp_info_buf.tcpi_retrans) RTT (var) is the
sampled RTT variance. All stats taken from tcp_info or tcp_connection_info and sampled per the -i
reportting interval
NetPwr (*nix only) Network power defined as (throughput / RTT)
iperf -c host.domain.com -i 1 --bounceback --permit-key=mytest --hide-ips
------------------------------------------------------------
Client connecting to (**hidden**), TCP port 5001
Bursting: 100 Byte writes 10 times every 1.00 second(s)
Bounce-back test (size= 100 Byte) (server hold req=0 usecs)
TCP window size: 16.0 KByte (default)
------------------------------------------------------------
[mytest(1)] local *.*.*.96 port 38044 connected with *.*.*.123 port 5001 (bb len/hold=100/0)
(icwnd/mss/irtt=14/1448/10605)
[ ID] Interval Transfer Bandwidth BB cnt=avg/min/max/stdev Rtry
Cwnd/RTT RPS
[mytest(1)] 0.00-1.00 sec 1.95 KBytes 16.0 Kbits/sec 10=11.949/9.662/19.597/3.127 ms 0
14K/10930 us 83 rps
[mytest(1)] 1.00-2.00 sec 1.95 KBytes 16.0 Kbits/sec 10=10.004/9.651/10.322/0.232 ms 0
14K/10244 us 99 rps
[mytest(1)] 2.00-3.00 sec 1.95 KBytes 16.0 Kbits/sec 10=10.582/9.720/14.831/1.573 ms 0
14K/10352 us 94 rps
[mytest(1)] 3.00-4.00 sec 1.95 KBytes 16.0 Kbits/sec 10=11.303/9.940/15.114/2.026 ms 0
14K/10832 us 88 rps
[mytest(1)] 4.00-5.00 sec 1.95 KBytes 16.0 Kbits/sec 10=11.148/9.671/14.803/1.837 ms 0
14K/10858 us 89 rps
[mytest(1)] 5.00-6.00 sec 1.95 KBytes 16.0 Kbits/sec 10=10.207/9.695/10.729/0.356 ms 0
14K/10390 us 97 rps
[mytest(1)] 6.00-7.00 sec 1.95 KBytes 16.0 Kbits/sec 10=10.871/9.770/14.387/1.547 ms 0
14K/10660 us 91 rps
[mytest(1)] 7.00-8.00 sec 1.95 KBytes 16.0 Kbits/sec 10=11.224/9.760/14.993/1.837 ms 0
14K/11027 us 89 rps
[mytest(1)] 8.00-9.00 sec 1.95 KBytes 16.0 Kbits/sec 10=10.719/9.887/14.553/1.455 ms 0
14K/10620 us 93 rps
[mytest(1)] 9.00-10.00 sec 1.95 KBytes 16.0 Kbits/sec 10=10.775/9.689/14.746/1.562 ms 0
14K/10596 us 92 rps
[mytest(1)] 0.00-10.02 sec 19.5 KBytes 16.0 Kbits/sec 100=10.878/9.651/19.597/1.743 ms 0
14K/11676 us 91 rps
[ 1] 0.00-10.02 sec BB8(f)-PDF: bin(w=100us): cnt(100)=97: 5,98: 8,99: 10,100: 8,101: 12,102:
10,103: 6,104: 7,105: 2,106: 2,107: 3,108: 3,109: 2,110: 1,114: 1,115: 1,118: 1,120: 2,121: 1,124:
1,125: 1,128: 1,140: 1,143: 1,144: 1,146: 2,148: 1,149: 2,150: 1,151: 1,152: 1,196: 1
(5.00/95.00/99.7%=97/149/196,Outliers=0,obl/obu=0/0)
where BB cnt=avg/min/max/stdev Count of bouncebacks, average time, minimum time, maximum time, standard
deviation units of ms
Rtry Total number of TCP retries
Cwnd/RTT (*nix only) TCP congestion window and round trip time (sampled where NA indicates no
value)
RPS Responses per second
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
where (per -e,)
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
where (per -e,)
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
where (per -e,)
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
where (per -e,)
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
where (per -e,)
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)
where, 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
Note: 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.
IP tos and dscp: Specifies the type-of-service or DSCP class for connections. Accepted values are af11,
af12, af13, af21, af22, af23, af31, af32, af33, af41, af42, af43, cs0, cs1, cs2, cs3, cs4, cs5, cs6, cs7,
ef, le, nqb, nqb2, ac_be, ac_bk, ac_vi, ac_vo, lowdelay, throughput, reliability, a numeric value, or
none to use the operating system default. The ac_xx values are the four access categories defined in WMM
for Wi-Fi, and they are aliases for DSCP values that will be mapped to the corresponding ACs under the
assumption that the device uses the DSCP-to-UP mapping table specified in IETF RFC 8325.
One can set the tos byte using:
--dscp on the client
--tos or -S on the client or server though the server side setting has caveats (per next item)
--tos-override for server side reversed traffic (use case is for bleaching the return path)
The --dscp value on the client is merely a convenience in that the user can set the dscp field's six bits
vs set the tos byte. The --tos-override allows one to test the reflected TOS feature of APs. This is
when the network bleaches the client side tos setting and the reverse traffic will then have no TOS
setting. The AP is expected to apply the upstream traffic's TOS to the reverse traffic. The use case is
that the --tos-overide would by zero simulating network bleaching of the tos byte. If there is no --tos
setting then the setsockopt will never be called and the default values are used.
--trip-times The --trip-times option enables many one way delay (OWD) metrics. Also note that using
--trip-times on a TCP client will cause --tcp-write-prefetch to be set to a small value if
tcp-write-prefetch hasn't hasn't also been set. This is done to reduce send side bloat latency (which is
unrelated to network induced latency.) Set --tcp-write-prefetch to zero to disable this (which will
disable TCP_NOTSENT_LOWAT) and will allow for send side bloat.
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. See bounceback NOTES section on clock unsynchronized
detections
Histograms and non-parametric statistics: 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, minimum, maximum and variation. This loses information when the
underlining distribution is not Gaussian. Histograms are supported so this information is made
available.
Histogram output interpretation: Below is an example bounceback histogram and how to interpret it
[ 1] 0.00-5.10 sec BB8-PDF:
bin(w=100us): cnt(50)=35: 1,37: 1,39: 1,40: 3,41: 4,42: 1,43: 1,52: 1,57: 1,65: 1,68: 1,69: 1,70: 1,72:
2,74: 1,75: 5,78: 1,79: 2,80: 4,81: 3,82: 1,83: 1,88: 2,90: 2,92: 1,94: 1,117: 1,126: 1,369: 1,1000:
1,1922: 1,3710: 1 (5.00/95.00/99.7%=39/1000/3710,Outliers=4,obl/obu=0/0)
where, [ 1] The traffic thread number
0.00-5.10 sec The time interval of the histogram
BB8-PDF BB8 is the histogram name and the PDF indicates a histogram raw output
bin(w=100us) provides the bin width. The bin width of this histogram is 100 microseconds
cnt(50) provides the total number of samples in the histogram. There are 50 samples in this
histogram
35:1 provides the bin no then the number of samples in that bin. Bin 35 with bin width 100us is
3.4 ms - 3.5 ms and there was one sample that landed there
5.00/95.00/99.7%=39/1000/3710 provides the bin confidence intervals (per the integrated cumulative
distribution function.) 5% landed in 3.9 ms or better (recall bin number multiplies by bin width.)
95% landed in 10 ms or better. 99.7% or 3 standards of deviation landed in 37.1 ms or better
Outliers=4 provides the outlier count, similar to 3IQR (3 times the inter quartile range) but uses
10% and 90% for inner & outer fence post, then 3 times that for outlier detection.
obl/obu=0/0 out of bounds lower and out of bands upper, provides the number of samples that could
not be binned because the value landed outside of all possible bins
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.) The units are responses per second or rps. 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. Results are shown in units of rps and time
delays.
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 have TCP run in default mode per
the socket (which is most likely TCP_QUICKACK being off.)
Unsynchronized clock detections with --bounceback and --trip-times (as of March 19, 2023): Iperf 2 can
detect when the clocks have synchronization errors larger than the bounceback RTT. This is done via the
client's send timestamp (clock A), the server's receive timestamp (clock B) and the client's final
receive timestamp (clock A.) The check, done on each bounceback, is write(A) < read(B) < read(A). This is
supported in bounceback tests with a slight adjustment: clock write(A) < clock read(B) < clock read(A) -
(clock write(B) - clock read(B)). All the timestamps are sampled on the initial write or read (not the
completion of.) Error output looks as shown below and there is no output for a zero value.
[ 1] 0.00-10.00 sec Clock sync error count = 100
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_QUICKACK: The TCP_QUICKACK socket option will be applied after every read() on the server such that
TCP acks are sent immediately, rather than possibly delayed.
TCP_TX_DELAY (--tcp-tx-delay): Iperf 2 flows can set different delays, simulating real world conditions.
Units is microseconds. This requires FQ packet scheduler or a EDT-enabled NIC. Note that FQ packet
scheduler limits might need some tweaking
man tc-fq
PARAMETERS
limit
Hard limit on the real queue size. When this limit is
reached, new packets are dropped. If the value is lowered,
packets are dropped so that the new limit is met. Default
is 10000 packets.
flow_limit
Hard limit on the maximum number of packets queued per
flow. Default value is 100.
Use of TCP_TX_DELAY option will increase number of skbs in FQ qdisc, so packets would be dropped if any
of the previous limit is hit. Using big delays might very well trigger old bugs in TSO auto defer logic
and/or sndbuf limited detection.
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)
NLANR/DAST April 2024 IPERF(1)