Interesting observation on bbr v3 kernel and ss -tin

99 views
Skip to first unread message

muhamm...@umt.edu.pk

unread,
Aug 1, 2025, 12:35:57 PMAug 1
to bbr...@googlegroups.com

Hi there,

 

I have an interesting observation to share and ask, while doing testing on BBR v3 based compiled kernel.

 

I was doing upload tests on a wired ethernet server ( with 250-300Mbps bandwidth available) using ethernet adapter. I enabled  BBRv3 as the default congestion control via linux sysctl.

 

Ss -tin was used to observe the output , when the upload started for firsts few seconds ss -tin  reported Cubic in the output  and then shifted to showing bbr.

 

At home, I tried the same test again but now using WIFI adapter and made bbr v3 the default using sysctl and ran the upload test and used ss -tin to observe.

 

Ss -tin  at once started to show BBR and not a second it showed cubic.

 

 

My question is why using ethernet adapter on a wired network , the ss -tin output always shows cubic for first few seconds and then transitions to bbr.

 

 

(I am using  Linux Kernel 6.13.7 with bbr v3 compiled in it via github instructions from Neal Cardwell.)

 

 

 

Regards,

Ahsan

 

 

 

Neal Cardwell

unread,
Aug 1, 2025, 12:48:09 PMAug 1
to muhamm...@umt.edu.pk, bbr...@googlegroups.com
That's quite surprising. Are you sure you are looking at the correct "ss" output lines? :-) Often networking benchmark tools have a "control" connection and a "data" connection, and perhaps what you are seeing is that the "control" connection is using cubic but the "data" connection is using bbr?

(1) What tool are you using for testing?

(2) What are the exact command lines you are using for your test?

(2) Can you please attach a log file demonstrating the odd behavior? From something like the following:

(while true; do date; ss -tenmoi; sleep 0.5; done) > /root/ss.txt &

Thanks,
neal


--
You received this message because you are subscribed to the Google Groups "BBR Development" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bbr-dev+u...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/bbr-dev/000201dc0302%2455ba6890%24012f39b0%24%40umt.edu.pk.

muhamm...@umt.edu.pk

unread,
Aug 5, 2025, 3:17:20 AMAug 5
to Neal Cardwell, bbr...@googlegroups.com

 

 

 

Regards,

 

 

From: Neal Cardwell <ncar...@google.com>
Sent: Friday, August 1, 2025 9:48 PM
To: muhamm...@umt.edu.pk
Cc: bbr...@googlegroups.com
Subject: Re: [bbr-dev] Interesting observation on bbr v3 kernel and ss -tin

 

That's quite surprising. Are you sure you are looking at the correct "ss" output lines? :-) Often networking benchmark tools have a "control" connection and a "data" connection, and perhaps what you are seeing is that the "control" connection is using cubic but the "data" connection is using bbr?

(1) What tool are you using for testing?

 

Flent

 

(2) What are the exact command lines you are using for your test?

 

Using flent command line to run upload test on a wired remote server on LAN.

 

Running vmware workstation with kernel “Linux ubuntu 6.13.7+v3+90210de4b779+GCE #1 SMP PREEMPT_DYNAMIC Sun Mar 23 13:11:21 PKT 2025 x86_64 x86_64 x86_64 GNU/Linux”

 

 


(2) Can you please attach a log file demonstrating the odd behavior? From something like the following:

(while true; do date; ss -tenmoi; sleep 0.5; done) > /root/ss.txt &

attached

 


Thanks,
neal

 

ss-bbr.txt
ss.txt

Neal Cardwell

unread,
Aug 5, 2025, 10:03:59 AMAug 5
to muhamm...@umt.edu.pk, bbr...@googlegroups.com
Hi,

From the traces you provided, it looks like all connections that used BBR used BBR for their entire lifetime, which makes sense.

If you still suspect there is a problem, can you please paste in some lines in a trace file and describe why you feel it shows a problem?

Thanks,
neal

ps: details:

$ grep -B1 bbr ss.txt | grep -v cwnd | awk '{print $4, $5}' | sort | uniq -c
 499  
 103 192.168.208.130:36793 10.103.108.13:46549
 103 192.168.208.130:39649 10.103.108.13:12865
 500 192.168.208.130:51266 108.177.15.84:443
   1 192.168.208.130%ens33:36436 91.189.91.49:80

$ grep -A1 "192.168.208.130:36793 10.103.108.13:46549" ss.txt | grep cubic | wc -l
       0
$ grep -A1 "192.168.208.130:39649 10.103.108.13:12865" ss.txt | grep cubic | wc -l
       0
$ grep -A1 "192.168.208.130:51266 108.177.15.84:443" ss.txt | grep cubic | wc -l
       0
$ grep -A1 "192.168.208.130%ens33:36436 91.189.91.49:80" ss.txt | grep cubic | wc -l
       0
---

$ grep -B1 bbr ss-bbr.txt | grep -v cwnd | awk '{print $4, $5}' | sort | uniq -c
 243  
 103 192.168.208.130:36793 10.103.108.13:46549
 103 192.168.208.130:39649 10.103.108.13:12865
 244 192.168.208.130:51266 108.177.15.84:443
$
$ grep -A1 "192.168.208.130:36793 10.103.108.13:46549" ss-bbr.txt | grep cubic | wc -l
       0
$ grep -A1 "192.168.208.130:51266 108.177.15.84:443" ss-bbr.txt | grep cubic | wc -l
       0



Reply all
Reply to author
Forward
0 new messages