Bufferbloat Test



  • I tried a Bufferbloat  test in dslreports  and got  F for Bufferbloat  on my upload ,  it spikes up to 400ms and over,  then I enabled codel gave it a little less on WAN and Lan but still got same results, I have a 100/6 connection, But what is strange is when I test myself  this way, I upload a file using full upload speed, downloading a torrents on 2 separate PCs, stream a 4k youtube video on a s6,open 2 terminals with constant ping to google.com and speedtest.net on a other phone, doing all this my minimun ping was 24ms once in a while goes to 45ms. so in  my own  test I am seeing good results, maybe I am testing wrong sure you guys will let me know :). I do not post this as a problem  just would like to maybe understand things better as here are great knowledgeable    guys  :) BTW stranger is when I give traffic shaper  half my upload speed  I still get F, I plug a PC direct to the cable modem with same results on dslreports but on my manual tests  pings went up to 500ms :))



  • The DSLReports test has accuracy problems with me as well. I just go by my own ping results along with pfSense's Quality graphs.

    I think it is because the latency is measured in-browser, but I dunno. All my PCs are slow Linux setups, so any in-browser calculations are resultingly slow. :-\



  • Something is not right in DSLREPORTS could b browser as you say, I used chrome on a win 10 PC  dual core 3.0GHZ 120gb Samsung ssd



  • What are you rules that you are using the shape and the queue setup?

    I've got pretty much the same and I've tested from Chrome on Mac/Linux and get A+ for bufferbloat.

    It took some tuning to figure out the right settings for the queues and the rules though.



  • @Animosity022:

    What are you rules that you are using the shape and the queue setup?

    I've got pretty much the same and I've tested from Chrome on Mac/Linux and get A+ for bufferbloat.

    It took some tuning to figure out the right settings for the queues and the rules though.

    I removed all shaper config and created just codel WAN  (upload) and LAN (download)  also tried running the wizard HFSC. If you got some config tips and do not mind sharing please do and will test.

    Regards



  • DSLReports does an HTTP ping via websockets. If you're doing a normal ICMP ping, then your traffic shaping may be placing ICMP into a different queue. Bufferbloat is a per-queue characteristic. Make sure you're measuring the same queue.



  • @Harvy66

    Thank you  for your  response (was hoping to get a response from you). I was doing a normal ICMP ping, ping google.com in cmd, so  should I test dsl reports again and see what q  is used when its testing? Thanks



  • pics No wizard  just codel




  • So for my config, I really do a pretty simple config as I have FIOS I'm on 300/300.

    I break up WAN/LAN into 3 queues (High/Med/Low) and do 100/100/90 for bandwidth respectively. I only use CODEL on each queue that is setup.

    My rules prioritize my ICMP and my XBox traffic. I deprioritize my 'other' plex server traffic and everything else falls in the default queue.

    I use 'quick' floating pass rules so I can see what states are captured in each rule and in each queue.

    I've tested a lot of downloads and uploads maxing each side of the connection. My biggest problem up to this point was not getting the size of the WAN/LAN config and really making sure I was at the minimum bandwidth I can expect and really a little lower. You sacrifice a bit of your pipe to ensure throughput.



  • @Chrismallia:

    pics No wizard  just codel

    Try without shaper and with shaper. Also try to limit bandwidth to the half of initial to find out is there a difference.
    I have similar shaper setup like Animosity022 have, but also using limiters to equalize bandwidth for all users.
    Bandwidth limits in shaper and limiters are set to 90% of real tested bandwidth, because I have had buffer bloat problem on ISP side, I have gigabit ethernet with 300/300 PPPoE and on Sunday evening it's common problem.



  • Hi w0w Great to hear from you. I tried with shaper also and also tried giving half my bandwidth speed but same results.



  • DSLreports is nuts  look where I limited my speeds  and just got this - for buferbloat  o and during the test it still spiked to 300ms




  • To make sure that your shaper works, also please check status -> queues, under heavy load you will see drops, Animosity022 showed it on screenshot. I think it works, just because your manual ping test shows it.
    As for dslreports, it looks like your bandwidth is 10 times less then you have said in the first post. I think there is the problem just with dslreports tests or some ISP limits that just don't let the test to run correctly, you can ask your ISP tech support.



  • The shaper works fine as I get drops and In my manual test turning off traffic shaping pings went up to 500ms turning back on shaper pings back down to 24ms, there is nothing wrong in pfsense in DSLReports there is something  funny, I called my isp and  we did some test they also told me that my pings are great under load they think that there is something wrong with the sites test or pings to the servers it is pinging. Also I tried a router with firmware DDWRT  with qos and same results, and something also is not right in my results I sometimes just get  ( - ) no letter  I also get this in the speed results



  • Well small update I tried the same test on the same ISP from a different location this guy uses the cisco combo modem provided by the ISP and he got grade  C on bufferbloat  also I ping google.com from a command prompt he gets 5ms ping I get 13ms. So same ISP different quality



  • Just arrived home and gave it another go and could not believe or understand what happened  look




  • I think ISP fixed something.


Log in to reply