ftp to dev null to test bandwidth

When testing bandwidth, and troubleshooting bottlenecks, I prefer to use iperf. If you insist on testing bandwidth with FTP, it’s important NOT to use regular files. If you transfer actual files, the transfer could be limited by disk i/o, due to reads and writes. To eliminate this you can FTP from /dev/zero to /dev/null. It sounds super easy, but you have to use FTP in a special way to get it to read and write to special devices.

Here’s a little script. Be sure to replace the destination IP address, username and password with actual values:

# cat ftp_dev_null.sh
/usr/bin/ftp -n <IP address of machine> <<END
verbose on
user <usernanme> <password>
put "|dd if=/dev/zero bs=32k" /dev/null
# ftp_dev_null.sh
Verbose mode on.
331 Password required for fordodone
230 User fordodone logged in
Remote system type is UNIX.
Using binary mode to transfer files.
200 Type set to I
local: |dd if=/dev/zero bs=32k remote: /dev/null
200 PORT command successful
150 Opening BINARY mode data connection for /dev/null
send aborted
waiting for remote to finish abort
129188+0 records in
129187+0 records out
4233199616 bytes (4.2 GB) copied, 145.851 s, 29.0 MB/s
226 Transfer complete
4233142272 bytes sent in 145.82 secs (28350.0 kB/s)
221 Goodbye.

In this case I was getting around 230Mbits per second (over an IPSec tunnel) between my client and the FTP server. Not too bad.

test bandwidth throughput using iperf

Testing bandwidth between 2 hosts can be helpful to determine maximum transfer rates, and how much bandwidth can be pushed. The iperf utility is a great option for this. Set up one side as a listening server and the other side as a client. In this scenario I want to test an OpenVPN tunnel between a Raspberry Pi and a Vyatta server. The layout looks like this:

Laptop --(wireless)--> RPi --(wireless)--> Linksys router -> Enterasys router -> Cable Modem -> ISP -> INTERNET -> Vyatta -> remote host

Start iperf in server mode on the remote host listening for UDP datagrams:

# iperf -s -u
Server listening on UDP port 5001
Receiving 1470 byte datagrams
UDP buffer size:   110 KByte (default)

Start up the client. The -c flag tells it what remote host to connect to with iperf, the -u makes it use UDP, the -b tells it what bandwidth to try to achieve, and the -t sets how many seconds to run for:

# iperf -c -u -b 1000K -t30

As the test runs, the server and client will output the rates. Here’s the server side output:

[  3] local port 5001 connected with port 40816
[ ID] Interval       Transfer     Bandwidth       Jitter   Lost/Total Datagrams
[  3]  0.0-30.0 sec  3.57 MBytes    998 Kbits/sec  14.738 ms    4/ 2552 (0.16%)

So we were able to push a full 1 Mbps and only lost 4 datagrams, that’s pretty good. I would say this can comfortably push 1Mbps. (Side note: top reported openvpn process on RPi was at 9% CPU utilization) Let’s try for 2Mbps:

# iperf -c -u -b 2000K -t30

And the output:

[  4] local port 5001 connected with port 56707
[  4]  0.0-30.0 sec  6.21 MBytes  1.74 Mbits/sec  7.346 ms  670/ 5103 (13%)

We were able to achieve 1.74 Mbps, but we lost 13% of the traffic. That is unacceptable! I would run again at 1500K and narrow in where exactly the break point is for throughput. To find the bottleneck you can run iperf from every (Linux) host in the chain to pinpoint the limitation. I suspect it could be the upload speed of my home internet connection, or the poor wireless signal of the client side of the RPi.