[tsvwg] Re: iperf2 Android update: Bounceback fixes and IPv6 support

[[e-mail]]

Hi All,

I've updated the iperf 2 android bounceback test to display the 
histogram with a CDF overlay. UDP L4S bounce back is still beind coded 
up and tested.

On 2026-07-22 11:57, [e-mail] wrote:

> Hi all,
> 
> An updated iperf2 Android build is available with two Bounceback fixes.
> 
> First, a bug that caused responsiveness and RPS tests under working 
> load to fail after a few seconds has been fixed. The update has been 
> verified with a 50-second Bounceback run under load.
> 
> Second, the selected congestion-control algorithm now applies at both 
> ends of the Bounceback connection. Anyone comparing congestion-control 
> algorithms should repeat those tests with this build.
> 
> The public responde.* servers support Reno, Cubic, and BBR, with Cubic 
> as the default. They do not currently support Prague. Prague testing 
> requires both a server and a device with Prague available.
> 
> The public servers have already been updated. If you use your own 
> iperf2 server, please update it from the 2-2-2 branch.
> 
> The public servers limit tests to 60 seconds, so please select a 
> shorter duration.
> 
> Android client presentation:
> 
> https://www.umbernetworks.com/IPPM_iperf2_Android_Pres.html
> 
> INSTALL THE UPDATED APK
> 
> Download this exact file from the SourceForge file listing:
> 
> iperf2.2.2_21.bb_cca_fix.July2026_android.apk
> 
> https://sourceforge.net/projects/iperf2/files/
> 
> Please select the filename directly rather than using the green 
> Download button, which may retrieve a different file. The APK can be 
> installed over an existing version.
> 
> FEATURES WORTH TESTING
> 
> Bounceback now displays the minimum RTT reported by the kernel's TCP 
> statistics, together with a theoretical RPS ceiling:
> 
> Theoretical RPS ceiling = 1,000 ms / minimum TCP RTT
> 
> For example, a minimum TCP RTT of 20 ms gives a theoretical ceiling of 
> 50 request-and-reply cycles per second.
> 
> Minimum RTT is a TCP statistic, while observed RPS is measured from the 
> application-level Bounceback exchanges. Because each request waits for 
> its reply before the next request is sent, actual RPS will normally be 
> lower because of application processing, scheduling, working load, and 
> other path effects.
> 
> The public responde.* servers in Fremont, Vienna, and Newark also 
> support IPv6. The app can select IPv6 normally through the Host field, 
> or you can enable Force IPv6 (-V) to test the IPv6 path explicitly. 
> Your local network must also support IPv6.
> 
> Thanks again to everyone testing the app and providing feedback. It is 
> very helpful.
> 
> Bob