iperf commands - for throughput / traffic

[jchellap]

Hi
With the open thread source code, can we use iperf tool and run througput or kind of tests?
Any suggestions for the same?

[jwhui]

OpenThread itself does not implement an iperf endpoint. However, given that iperf operates on top of IP, OpenThread can transport iperf message. For example, you could run iperf between to Linux hosts connected via Thread.

Alternatively, OpenThread does provide some related performance utilities - see tcp and ping CLI commands.

[jchellap]

@jwhui , Thanks! we are looking tools for throughput. However, given that iperf operates on top of IP, OpenThread can transport iperf message.
Does this mean I can use iperf command from the iperf tool or like CLI command should be used? Please quote example links which can be used to further understand on this. Thanks for the support

[Irving-cl]

@jchellap

I actually did this before. However it was not operated on universally availble devices like raspberry pi. But yes, it can run on your device (linux-based devices). I can show you the steps to run it:

• You need to first prepare 2 Thread devices (must be host+RCP mode). Because you need to run iperf on the linux system. Things like ot-cli-ftd, ot-ncp-ftd cannot run iperf.
• Setup the Thread network on your thread devices, make sure they are connected. You can ping to ensure this. Then you will see the wpan interface in the system of the two devices.
• You may need to turn off the firewall on your device. This depends on your device. For example, run this on both of your devices: ip6tables -I INPUT -p udp --dport 5001 -j ACCEPT
• Get the Thread IPv6 address on your receive end. I usually use the mleid address. You can get it ot-ctl: ot-ctl ipaddr mleid.
• Start the iperf server on your receiving end: iperf -V -s -u -i 1 -U. You may adjust the options based on your needs.
• Start the iperf client on your sending end: iperf -V -c {mleid} -t {time_to_run_in_seconds} -u -i 1 -b {bandwidth_kbps} -l {udp_packet_length}

This should work. But since there may be many differences between your devices and mine including the iperf version, you may need some debugging.

So for your question:

Does this mean I can use iperf command from the iperf tool or like CLI command should be used?

You should use iperf tool.

[jchellap]

One other question is what is the typical throughput value we can get for UDP using Open thread RCP design?

[Irving-cl]

One other question is what is the typical throughput value we can get for UDP using Open thread RCP design?

From what I got, you can usually get 60~70 Kbps for UDP throughput. We also achieved 80+(near 90) Kbps on some devices. Note that the performance also depends on the environment. You may want to use a RF-shielded space to run the test to achieve high throughput

[jchellap]

Thanks much for some reference data. This helps. I hope the measurement is based on the Server side for the RX right?

[jchellap]

What should be the bitrate at the client side? i used bit rate as 1K
iperf3 -c IPV6 address -V -u -p 5123 -i 1 -t 20 -b 1K -l 1200

[Irving-cl]

-b is the highest bandwidth you want. Iperf won't exceed the number. Also it may not reach the number. If you set 1K, it's 1Kbps at most. You can use 80K here. If it can reach 80K, you can try a larger value.

[jchellap]

Further one query, When i use Leader as server and Router as child and client, i see UDP messages going through between client and server...with some throughput values.
However when i start iperf3 on child as server and Leader as client, it doesnt work. No data exchange.
Is it not to be used? Any explanation for not working will help me understand the right method of usage.

[zhanglongxia]

I also met iperf3 issues in recent days. Here are the methods that I resolved issues. Hope it could help you.

1. IPerf3 will create a TCP connection before transmitting UDP messages. Thread stack has enabled the TCP feature by default. It causes iperf3 to be unable to successfully create a TCP connection. So we can disable the OPENTHREAD_CONFIG_TCP_ENABLE in Thread stack.
2. The SRP server will add the anycast address to the Thread network interface. The Linux network stack has a chance to select the anycast address as the source address of the Thread IP packet. Try to run the command ot-ctl srp server disable to resolve this issue.
3. On the iperf3 server side, bind a Thread network IP address. For example, iperf3 server: iperf3 -s -i 1 -B fd08:3400:dce1:8a22:5645:80f8:66a7:27f1, iperf3 client: iperf3 -6 -c fd08:3400:dce1:8a22:5645:80f8:66a7:27f1 -u -b 60k -i 1 -t 30.
4. Check the firewall on your devices to make sure that Thread IP packets are not blocked by the firewall.
5. Use the command ping6 on both devices to ping each other to make sure that the routing table is correct.
6. Use tcpdump -i wpan0 to monitor the traffic on the wpan0 interface to check whether the Thread IP traffic is as expected.

[jchellap]

Thanks will try the same.
I am able to successfully test when Leader is server and Child as Client.
I see no packet transmission/transaction when Leader is client and Child as Server. Is this usecase correct at all?

[zhanglongxia]

Iperf3 should work regardless of whether the Iperf3 server is leader or child. You can check the OpenThread log and the output of tcpdump -i wpan0 on both leader and child to find where the IP packet is dropped. That will help you find out the root cause.

[jchellap]

I tried disabling the OPENTHREAD_CONFIG_TCP_ENABLE in Thread and also ot-ctl srp server disable.
When i trigger iperf as client on Leader, i see some packets that is decrypted as TCP retransmission.

Anyone came across such issue?

[jchellap]

Just an update. With 2ce tag for ot-br-posix and open thread source code, i see that Leader configured as Client for UDP data transmission is not working and child device as server.
However, thread-reference-20221027 git commit, the child device as server is working as expected. Any particular commit that i can cherry-pick or should i move to the latest code base?

[jwhui]

Unfortunately, I'm not aware of any specific commit that would cause this issue. Would you be willing to do a git-bisect to pinpoint the commit?

[jchellap]

I will need sometime to check, the quick resolution I found was to move to later commit and continue. The other issues I found is crashing my tool due to extra literals used in the print in daemon.cpp and net_if.cpp.
In daemon.cpp
otLogInfoPlat("Session socket is ready"); --> this had additional literal such %s.
In net_if.cpp --> may be this is fixed already.. just for your info
otLogWarnPlat("[netif] Successfully added an external route %s in kernel:", prefixString); -> has extra %s

For the iperf.. will need sometime to find exactly which commit will caused the issue. Since there was some code changes in ip6.cpp interms of structure usage, i felt patching them will make it more hard to build the source code. The issue was on the receiving side of the iperf packet.

[jwhui]

The format string issues were fixed in #8353 and #8356.

[jchellap]

Thanks for sharing the above link. Had missed one formatting error in my copy.
In general how do we sync up such changes with the open source code? Can you recommend/suggest? Should we move to the latest tags etc? That is what I was following

[jchellap]

One other query, I saw your post in Linked about iphone hosting Open Thread.. In mobile phones, the device is going to be Border Router or What kind of roles it takes? Any link that can help me understand in real time use cases which type of nodes is selected?
For SSED kind of devices, which is the real time product that can be used?