Playing with fq_codel in 2.4



  • @Veldkornet I don't suppose it's a PPPoE connection? I had a similar problem.



  • Hi how did you resolve pppoe problem?
    Thanks!



  • @maverick_slo I didn't. I am only able to use the Download shaper for my PPPoE WAN. It works perfectly on my other WAN which is DHCP.



  • @csutcliff Nope, mine is just a DHCPv4 connection



  • @Veldkornet said in Playing with fq_codel in 2.4:

    @xciter327 Even at 95% of 400/40 which is 380/38, the upload speed is nowhere near that. The download at least comes in the vicinity.

    Hm. That should not be. Did You follow the guide I linked? Also what hardware is the pfsense running on? Limiters add extra CPU usage. Take a look with "htop -d 1" (need to install it first "pkg install htop") and see if You are peaking the CPU while doing the test.



  • @xciter327 ah! I didn't see the guide. I've now re-created everything as per the guide... although it didn't change the results...

    I have a PCEngines APU2. I just had a look in "top" while doing the tests. Didn't really see anything climb very high at all. Even then, if it can handle the download, it should be able to manage the upload which is 10% of the download.

    If I watch the test, the upload starts strong and climbs to around 20Mbit quickly, for 2 seconds or so, but then drops down to around 4/5 for the remainder of the test where it eventually finishes.



  • Could You perhaps post some picture of your firewall rules and limiter config? I am shaping my Ziggo connection on a Zotac NUC, which theoretically should be less powerful than the APU2. Also make sure You clear the states/reset the firewall when applying the limiters. Keep an eye in the system log for any log messages when You apply the limiters as well.



  • I can post screenshots instead of the below.. but it will take lots of space. I just checked everything though, and except for the speed limits, to me it looks the same as in the post. Might as well just be a copy paste. Looking at the floating rule screenshot though, I see that the WAN-In FQ-CoDel queue is pretty small considering all the tests I was doing... normal?

    Also, see how the upload just dies off:
    b13103a2-63f8-4d95-8ac4-fc2a13a21e21-image.png

    8668731a-351d-40d0-8f8d-fd51f481cfc4-image.png

    FQ_CODEL_OUT

    Name: FQ_CODEL_OUT
    Bandwidth: 38 Mbit/s
    Mask: None
    Queue Management Algorithm: Tail Drop
    Scheduler: FQ_CODEL
    target: 5
    interval: 100
    quantum: 300
    limit: 10240
    flows: 20480
    

    fq_codel_out_q

    Name: fq_codel_out_q
    Mask: None
    Queue Management Algorithm: Tail Drop
    

    FQ_CODEL_IN

    Name: FQ_CODEL_IN
    Bandwidth: 380 Mbit/s
    Mask: None
    Queue Management Algorithm: Tail Drop
    Scheduler: FQ_CODEL
    target: 5
    interval: 100
    quantum: 300
    limit: 10240
    flows: 20480
    

    fq_codel_in_q

    Name: fq_codel_in_q
    Mask: None
    Queue Management Algorithm: Tail Drop
    

    Firewall Rules - Floating:
    0cbe9dcf-3a46-4a2b-b529-90c1a1442677-image.png

    policy routing traceroute workaround

    Action: Pass
    Quick: Tick Apply the action immediately on match.
    Interface: WAN
    Direction: out
    Address Family: IPv4
    Protocol: ICMP
    ICMP subtypes: Traceroute
    Source: any
    Destination: any
    Description: policy routing traceroute workaround
    

    limiter drop echo-reply under load workaround

    Action: Pass
    Quick: Tick Apply the action immediately on match.
    Interface: WAN
    Direction: any
    Address Family: IPv4
    Protocol: ICMP
    ICMP subtypes: Echo reply, Echo Request
    Source: any
    Destination: any
    Description: limiter drop echo-reply under load workaround
    

    WAN-In FQ-CoDel queue

    Action: Match
    Interface: WAN
    Direction: in
    Address Family: IPv4
    Protocol: Any
    Source: any
    Destination: any
    Description: WAN-In FQ-CoDel queue
    Gateway: Default
    In / Out pipe: fq_codel_in_q / fq_codel_out_q
    

    WAN-Out FQ-CoDel queue

    Action: Match
    Interface: WAN
    Direction: out
    Address Family: IPv4
    Protocol: Any
    Source: any
    Destination: any
    Description: WAN-Out FQ-CoDel queue
    Gateway: WAN_DHCP
    In / Out pipe: fq_codel_out_q / fq_codel_in_q
    


  • Looks good to me. Mine at home is the same, with lower speeds tough. When You do the dslreports test, You can open up htop(prefer it because it's easier to deal with multiple cores) in one window and "ipfw sched show" in another to see if the limiters are actually matching traffic. Anything else on this box(like squid or snort)?



  • @xciter327 said in Playing with fq_codel in 2.4:

    Looks good to me. Mine at home is the same, with lower speeds tough. When You do the dslreports test, You can open up htop(prefer it because it's easier to deal with multiple cores) in one window and "ipfw sched show" in another to see if the limiters are actually matching traffic. Anything else on this box(like squid or snort)?

    Well I have Suricata, no squid. Although I turned Suricata off and it made no difference.

    Download maxes the CPU, but upload doesn't seem to do much...
    Download:

    00001:  38.000 Mbit/s    0 ms burst 0
    q65537  50 sl. 0 flows (1 buckets) sched 1 weight 0 lmax 0 pri 0 droptail
     sched 1 type FQ_CODEL flags 0x0 0 buckets 1 active
     FQ_CODEL target 5ms interval 100ms quantum 300 limit 10240 flows 20480 NoECN
       Children flowsets: 1
    BKT Prot ___Source IP/port____ ____Dest. IP/port____ Tot_pkt/bytes Pkt/Byte Drp
      0 ip           0.0.0.0/0             0.0.0.0/0     2478   112953 488 26508   0
    00002: 380.000 Mbit/s    0 ms burst 0
    q65538  50 sl. 0 flows (1 buckets) sched 2 weight 0 lmax 0 pri 0 droptail
     sched 2 type FQ_CODEL flags 0x0 0 buckets 1 active
     FQ_CODEL target 5ms interval 100ms quantum 300 limit 10240 flows 20480 NoECN
       Children flowsets: 2
      0 ip           0.0.0.0/0             0.0.0.0/0     33141 49232048 193 287300  12
    

    Capture.PNG
    Upload:

    00001:  38.000 Mbit/s    0 ms burst 0
    q65537  50 sl. 0 flows (1 buckets) sched 1 weight 0 lmax 0 pri 0 droptail
     sched 1 type FQ_CODEL flags 0x0 0 buckets 1 active
     FQ_CODEL target 5ms interval 100ms quantum 300 limit 10240 flows 20480 NoECN
       Children flowsets: 1
    BKT Prot ___Source IP/port____ ____Dest. IP/port____ Tot_pkt/bytes Pkt/Byte Drp
      0 ip           0.0.0.0/0             0.0.0.0/0       31    45860  0    0   0
    00002: 380.000 Mbit/s    0 ms burst 0
    q65538  50 sl. 0 flows (1 buckets) sched 2 weight 0 lmax 0 pri 0 droptail
     sched 2 type FQ_CODEL flags 0x0 0 buckets 1 active
     FQ_CODEL target 5ms interval 100ms quantum 300 limit 10240 flows 20480 NoECN
       Children flowsets: 2
      0 ip           0.0.0.0/0             0.0.0.0/0       16      664  0    0   0
    

    Capture2.PNG



  • Hm, I don't think You will have a great experience with all those thing's You've loaded on that little box. I read somewhere the APU2 is good for ~400Mpbs without Suricata/other heavy software.

    I would try disabling all add-on(haProxy, Openvpn, pfblocker, Suricata, snmp, fancy unbound settings, tftp-server etc.) and try vanilla pfsense with just the limiters(via floating rules) and a simple "Allow all" on the LAN side. If your CPU is peaked(like You have on the download test), then better run without limiters.



  • Oh? The CPU load on it is almost non-existant usually. This is the first time I've seen it go so high now with the traffic shaping. Even so, I can understand that the load may be an issue for download and I'm okay with that.

    It's the upload speed that's annoying me...... which is not doing much on the CPU side.



  • Well If I look at the picture, the upload is mostly pegged by Surricata. So that would be the first thing I disable. Also note that not all packages work normally with limiters. There used to be issues with Suricata/Squid and limiters, I don't know if it was ever fixed.



  • Okay, I disabled all of the packages, ran the test, upload good.
    Then I enabled each one and tested until the upload speed decreased.

    So it seems I have discovered the culprit... I have an OpenVPN Client setup to PIA, this is what's causing the bad upload speeds. I have an alias with an IP range defined, and basically everything within that alias should go over the VPN. And then I have an interface & gateway which is bound to the OpenVPN client; with a Firewall rule which says anything to that alias should use the VPN gateway.

    Is there anything special that I would need to do with my rules in this situation? The traffic itself isn’t going over the VPN.



  • My guess would be to place it at the very top of the floating rule set. I believe pf does "match most specific" unless "quick match" is selected.



  • @xciter327 said in Playing with fq_codel in 2.4:

    My guess would be to place it at the very top of the floating rule set. I believe pf does "match most specific" unless "quick match" is selected.

    pf is last match wins unless quick is enabled. Quick is enabled by default on interface rules but has to be selected on floating rules



  • Good to hear. I did not know about the default quick on interface rules. That explains some things.



  • I've rebuilt my network a bit, but I am not sure when this was happened — one PC behind pfSense always get the A or A+ rating and the other get the D or even lower grade rating on dslreports, both are on the same LAN. FQ_CODEL configuration is similar to the others in this thread except only that it is applied on the LAN side (LAN to any). When I disable limiters, then both PCs get A or A+ rating. Does anyone have an idea, what is going wrong?



  • What about Dual WAN?

    I've used the previous post (https://forum.netgate.com/post/807490), with my best guess modifications for two WANs (duplicated rules for each WAN, and duplicated queues), and received good grades on the DSLReports test. However, I use a dual WAN fallover setup. I noticed, today, when WAN1 went down and PFSense switched to WAN2, all traffic stopped. Removing the floating rules fixed the issue.

    The two WANs have different speeds.

    Does anyone have this working with dual WAN setup (different WAN speeds) in a fallover configuration? If so, what do your floating rules look like?

    @uptownVagrant Do you have any suggestions to modify your steps for dual WAN w/ fallover?

    Thanks, in advance, for any suggestions.



  • OK! I have fallen down the rabbit hole of bufferbloat! EEK!
    I've followed the Netgate video/slides on creating limiters without much benefit.
    I get a B for bufferbloat at DSLReports and A's for Quality and Overall.
    I have a 300/25 (advertised) Comcast cable connection. Tried changing the bandwidth down (reducing) , and the queue lengths (+/-/eliminating) w/o benefit. At this point, the DL queue length is set at 2000 and seems to be best so far. UL does not appear to be a problem. Any thing I can do short of getting a life? :)



  • @provels without the limiter rules enabled, and your filters reloaded and old states killed, what speeds are you seeing when you run the dslreports.com/speedtest for 60 seconds? It could be that Comcast's burst capacity during short tests may be heavily skewing your results - I seem to remember the default test duration only being 20 seconds. I think you may have to create an account and go into the preferences of the speed test to increase upload and download duration to 60s. There's always the flent project if you're savvy.

    Another thought, depending on how close your current limiter config is to the Comcast circuit sustained cap, you may be running into some buffering on your cable modem. What DOCSIS version is your cable modem operating?

    I'm assuming you're testing with a computer attached to your router/switch with an Ethernet cable and not using Wi-Fi.

    You could try halving your advertised values in your limiters and then increasing from there after each test. Note, if you're using FQ_CoDel then the queue length that you set in the limiter config is not being used. FQ_CoDel manages queue lengths dynamically and works to dequeue packets within the target you set.

    Here's a guide that might help although I would suggest leaving target, interval, quantum, limit, and flows set as the pfSense defaults for the initial tests.:
    https://forum.netgate.com/topic/112527/playing-with-fq_codel-in-2-4/815

    Best of luck



  • @ohbobva I'm running dual WAN with different speeds, and all you need to do is duplicate your setup. i.e. create four different limiters - upstream WAN1, upstream WAN2, downstream WAN1, downstream WAN2, and then a queue within each of those limiters. Set up four floating rules for WAN1-IPv4, WAN1-IPv6, WAN2-IPv4, WAN2-IPv6 and it all works as advertised. Be careful not to just create two queues within the same limiter as the limit will be the aggregate bandwidth of the two.



  • @Pentangle said in Playing with fq_codel in 2.4:

    @ohbobva I'm running dual WAN with different speeds, and all you need to do is duplicate your setup. i.e. create four different limiters - upstream WAN1, upstream WAN2, downstream WAN1, downstream WAN2, and then a queue within each of those limiters. Set up four floating rules for WAN1-IPv4, WAN1-IPv6, WAN2-IPv4, WAN2-IPv6 and it all works as advertised. Be careful not to just create two queues within the same limiter as the limit will be the aggregate bandwidth of the two.

    I'm having hard time to follow you, do you mind taking a screenshot of your config ? Thanks in advance, very much appreciated.



  • @uptownVagrant said in Playing with fq_codel in 2.4:

    Thanks very much for your quick reply ;)

    @provels without the limiter rules enabled, and your filters reloaded and old states killed, what speeds are you seeing when you run the dslreports.com/speedtest for 60 seconds? It could be that Comcast's burst capacity during short tests may be heavily skewing your results - I seem to remember the default test duration only being 20 seconds. I think you may have to create an account and go into the preferences of the speed test to increase upload and download duration to 60s. There's always the flent project if you're savvy.

    Did as you suggested, I get results as follows. Looked at flent, but I'm Windows. My upload is no issue, with or without limiters, but the download bloat is normally in the 60-90ms range.

    179/24 Mbps
    189/23
    169/24
    

    This result is pretty typical.
    06669f6b-a55e-4d35-ae2b-d378e2aedb38-image.png
    This is a typical result using the Ookla Speed Test (client version)
    2ff55461-d592-4813-80df-78548cf86bb0-image.png >

    Another thought, depending on how close your current limiter config is to the Comcast circuit sustained cap, you may be running into some buffering on your cable modem. What DOCSIS version is your cable modem operating?

    Modem is a Netgear CM600, DOCSIS 3.0, 24x8, Broadcom

    I'm assuming you're testing with a computer attached to your router/switch with an Ethernet cable and not using Wi-Fi.

    Yes, testing wired, but my pfSense is a VM running in Hyper-V and I'm testing from the host Desktop session. Maybe I'll wire up my laptop.

    You could try halving your advertised values in your limiters and then increasing from there after each test.

    I'll start by halving my DL BW and add from there.

    Note, if you're using FQ_CoDel then the queue length that you set in the limiter config is not being used. FQ_CoDel manages queue lengths dynamically and works to dequeue packets within the target you set.

    Noted and deleted.

    Here's a guide that might help although I would suggest leaving target, interval, quantum, limit, and flows set as the pfSense defaults for the initial tests.:
    https://forum.netgate.com/topic/112527/playing-with-fq_codel-in-2-4/815

    Best of luck

    Thanks for the tips. I'll check out the guide and report back. Thanks again! :)



  • I can start a new thread if more appropriate...

    What is the best way to prioritize WiFi Calling traffic while using an FQ_CoDel limiter setup?

    I have a setup with 30 - 80 WiFi clients. Currently using limiters and FQ_CoDel which seems to share the bandwidth very nicely. We have been having problems with WiFi Calling not being super reliable. We are in a cellular dead zone so people are relying on it.

    I was able to improve WiFi calling reliability by changing the Firewall Optimization to Conservative and changing the outbound NAT mode to Automatic rule generation.

    I would like to further optimize by making sure all the WiFi calling traffic has top priority.



  • This really is the thread from hell isn't it? I'd do a new thread.

    Over on linux (and not bsd as yet), we did this:

    https://www.usenix.org/system/files/conference/atc17/atc17-hoiland-jorgensen.pdf

    Skip to the MOS score at the end, and work back. There's support now for intel, qca, and mediatek chips. However... if you can get the clients to dscp mark for the VO or VI queue for how your AP defines it, that helps in that direction, and
    it is generally possible to build a more complicated qos/sqm setup that explicitly prioritizes voip out of the ipfw tools.

    In general I'm a big believer in short (fair) queues and lots of 'em, and not huge on prioritization. sch_cake (also mentioned on this thread), has some built-in optimizations as well.



  • This post is deleted!


  • @dtaht said in Playing with fq_codel in 2.4:

    This really is the thread from hell isn't it? I'd do a new thread.

    Over on linux (and not bsd as yet), we did this:

    https://www.usenix.org/system/files/conference/atc17/atc17-hoiland-jorgensen.pdf

    Skip to the MOS score at the end, and work back. There's support now for intel, qca, and mediatek chips. However... if you can get the clients to dscp mark for the VO or VI queue for how your AP defines it, that helps in that direction, and
    it is generally possible to build a more complicated qos/sqm setup that explicitly prioritizes voip out of the ipfw tools.

    In general I'm a big believer in short (fair) queues and lots of 'em, and not huge on prioritization. sch_cake (also mentioned on this thread), has some built-in optimizations as well.

    Thank you, dthat. I will look at this information.

    New thread:
    https://forum.netgate.com/topic/145924/prioritize-wifi-calling-traffic-and-fq_codel



  • After limiters and rules are in effect and states have been initially cleared, is it necessary to dump the states every time a modification is made to the limiters during testing? Thanks.



  • @provels said in Playing with fq_codel in 2.4:

    After limiters and rules are in effect and states have been initially cleared, is it necessary to dump the states every time a modification is made to the limiters during testing? Thanks.

    I clear them every time. Also monitor syslog to check if there are any errors.



  • I keep seeing these errors in my system log-

    fq_codel_enqueue maxidx = 797
    fq_codel_enqueue over limit
    fq_codel_enqueue maxidx = 797
    fq_codel_enqueue over limit
    fq_codel_enqueue maxidx = 797
    fq_codel_enqueue over limit
    fq_codel_enqueue maxidx = 797
    fq_codel_enqueue over limit
    fq_codel_enqueue maxidx = 797
    

    They seem to show at random times maybe 2 or 3 times a week and will repeat for a few minutes before stopping. Not sure if they represent a problem or how to fix them? Could definitely use any advice.



  • Hi guys, I'm new here and just followed the YouTube video from Netgate on how to set up fq codel with limiters. But then browsing this thread, I see people saying those settings are wrong and to use 'Tail Drop' as the queue management algorithm? I'm confused, why?

    Also, any recommendation on settings I should use? I have a gigabit verizon fios connection. It's hit 940/900 on Verizon's test if I recall correctly. Actually the upload has gotten over 900 a few times, the download is usually 500-700 on normal speed test sites except Verizon's where it can hit 900+ (and dslreports which has gotten it to 800+ a few times). I have a Plex Media Server and a gaming machine on the network.

    Using the settings from the YouTube vid got my Bufferbloat score to go from A to A+ in dslreports' test.



  • @subzerogts If you are at A+, I don't expect you have much to gain, LOL. I followed the video and also tried the tail drop config. Nothing seems to help me. Bloat is a B/C, Quality is A/B and Speed is D/F, even though the Ookla speedtest client gets me my advertised 300/25 to various local servers (even when limiting the Down limiter to 150. Nonsensical. Completely random speedtest-cli results testing from an inside server through pfSense and at the DSLReports test page. I don't know if maybe it's Comcast, my modem or what.
    ddb1039d-0116-4bd4-8046-0d6ee00ab863-image.png
    e0e0a0ac-6e24-472a-9da4-97854c1953af-image.png



  • If I want to disable the limiters, I know there are checkboxes for that. But if I just disable the floating rules and reset states, that gets them out of the system anyway, right?



  • @wgstarks said in Playing with fq_codel in 2.4:

    I keep seeing these errors in my system log-

    fq_codel_enqueue maxidx = 797
    fq_codel_enqueue over limit
    fq_codel_enqueue maxidx = 797
    fq_codel_enqueue over limit
    fq_codel_enqueue maxidx = 797
    fq_codel_enqueue over limit
    fq_codel_enqueue maxidx = 797
    fq_codel_enqueue over limit
    fq_codel_enqueue maxidx = 797
    

    They seem to show at random times maybe 2 or 3 times a week and will repeat for a few minutes before stopping. Not sure if they represent a problem or how to fix them? Could definitely use any advice.

    @wgstarks - what are your fq-codel parameters set to? One thing you might try is increasing the the value for the limit parameter. Here is a a link to some good documentation on what each parameter does:

    http://caia.swin.edu.au/freebsd/aqm/downloads.html

    Hope this helps.



  • @provels said in Playing with fq_codel in 2.4:

    @subzerogts If you are at A+, I don't expect you have much to gain, LOL. I followed the video and also tried the tail drop config. Nothing seems to help me. Bloat is a B/C, Quality is A/B and Speed is D/F, even though the Ookla speedtest client gets me my advertised 300/25 to various local servers (even when limiting the Down limiter to 150. Nonsensical. Completely random speedtest-cli results testing from an inside server through pfSense and at the DSLReports test page. I don't know if maybe it's Comcast, my modem or what.
    ddb1039d-0116-4bd4-8046-0d6ee00ab863-image.png
    e0e0a0ac-6e24-472a-9da4-97854c1953af-image.png

    @provels - how do the results change if you reduce the number of parallel streams during the DSL Reports speed test? Also, have you tried changing / tuning any of the fq-codel parameters?

    Something else I thought of: I admit that I don't know a lot about cable modems, but is buying a DOCSIS 3.1 cable modem an option? As far as I can tell DOCSIS 3.1 includes AQM part of the specification:

    https://tools.ietf.org/html/rfc8034

    Hope this helps.



  • @subzerogts said in Playing with fq_codel in 2.4:

    Hi guys, I'm new here and just followed the YouTube video from Netgate on how to set up fq codel with limiters. But then browsing this thread, I see people saying those settings are wrong and to use 'Tail Drop' as the queue management algorithm? I'm confused, why?

    Also, any recommendation on settings I should use? I have a gigabit verizon fios connection. It's hit 940/900 on Verizon's test if I recall correctly. Actually the upload has gotten over 900 a few times, the download is usually 500-700 on normal speed test sites except Verizon's where it can hit 900+ (and dslreports which has gotten it to 800+ a few times). I have a Plex Media Server and a gaming machine on the network.

    Using the settings from the YouTube vid got my Bufferbloat score to go from A to A+ in dslreports' test.

    When you select FQ_CODEL as scheduler you don't have to worry about setting up queue management algorithms for the queues because fq-codel manages per flow created packet queues automatically as part of fq-codel algorithm. This is why you don't really see any difference in performance if you e.g. select Codel as the queue management algorithm if you have FQ_CODEL selected.

    Having said that, fq-codel is a bit unique in this regard (in that it also manages per flow queues vs. just scheduling packets). In a way you can kind of think of fq-codel as a hybrid or all-in one packet scheduler and queue management algorithm. However, this is not the case for some of the other algorithms you have available to choose from as well as schedulers (e.g. QFQ, Round Robin, etc.). These are actually just scheduling algorithms that determine how packets should be dequeued from one more packet queues (e.g. Round Robin, weighted, etc.) When using one of these scheduling-only algorithms, you'll still have to setup packet queues and then choose an algorithm to manage those queues. In such as scenario, choosing a queue management algorithm will be important and will make a difference in performance.

    Hope this helps.



  • @tman222 BB goes to A and Speed tanks when I reduced DL streams from 24 to 6 or 12. Already bought the new Netgear modem after Comcast bitched about my Motorola for 2 years. Screw it. I'll monitor thread, but for now, I'm off the program.
    Nbbe627a8-720e-48e1-87a4-390519e414ca-image.png



  • @provels said in Playing with fq_codel in 2.4:

    @tman222 BB goes to A and Speed tanks when I reduced DL streams from 24 to 6 or 12. Already bought the new Netgear modem after Comcast bitched about my Motorola for 2 years. Screw it. I'll monitor thread, but for now, I'm off the program.
    Nbbe627a8-720e-48e1-87a4-390519e414ca-image.png

    Thanks @provels for getting back to me. If you don't mind me asking, how are you testing? Are the results consistent across different machines and browsers?



  • @tman222 My pfSense is a VM on Hyper-V and I'm testing from the host through the 10Gb Hyper-V interfaces through an Intel I340 Gb card hosting the v-switches via Cat6 to the modem. It's the only wired machine I have and no diff with other browsers. Results are completely random. Thanks for the help, but I'm not thinking it's worth the effort.


Log in to reply