Playing with fq_codel in 2.4



  • Just for testing purpose, try to change bw limiting to delay limiting :

    pipe 1 config delay 0ms
    

    for both pipes



  • Sorry gone for a few days, vacation.

    I gave that a shot by changing the /tmp file, it doesn't seem to have an affect. I am only changing the /tmp file, maybe it needs to be rebooted and hardcoded into the file? The only reason I haven't done this is because I haven't seen the results everybody is reporting…



  • Yes it's need to be rebooted or reloaded with```
    /etc/rc.reload_all

    After you did that run the following command```
    ipfw sched show
    ```and you should see something like```
    00001: unlimited         0 ms burst 0
    ```for the both pipes you have.


  • @w0w:

    Yes it's need to be rebooted or reloaded with```
    /etc/rc.reload_all

    After you did that run the following command```
    ipfw sched show
    ```and you should see something like```
    00001: unlimited         0 ms burst 0
    ```for the both pipes you have.
    

    Yes, all relatively simple and you've been great at walking through the steps you put in place.

    I'm getting this for both pipes.

    00003: unlimited        0 ms burst 0

    00004: unlimited        0 ms burst 0

    Yet I can't get better than a B rating for bufferbloat, which is the same if I literally do nothing at all….



  • But what about VPN bandwidth? Are you still getting 120Mbps?



  • That part HAS improved, looks like it does get about 145-ish or so which is about right. It just does nothing for bufferbloat.



  • Can you post the full output of```
    ipfw sched show



  • I'd like to look at implementing this, but I was wondering

    Anyone know the status of pfsync + limiters?



  • @moscato359:

    I'd like to look at implementing this, but I was wondering

    Anyone know the status of pfsync + limiters?

    What was the last status you know?  :D



  • The last status I know is that the pfsense book says not to use pfsync and limiters together, but doesn't explain why



  • @moscato359:

    The last status I know is that the pfsense book says not to use pfsync and limiters together, but doesn't explain why

    This is actual. https://redmine.pfsense.org/issues/4310 have 0% progress.



  • @w0w:

    @moscato359:

    The last status I know is that the pfsense book says not to use pfsync and limiters together, but doesn't explain why

    This is actual. https://redmine.pfsense.org/issues/4310 have 0% progress.

    D=



  • Is there any chance fq_codel will make it into the 2.4 GUI in limiters?



  • Definitely not!
    They are keeping eyes on it, but currently no plans, no moves, AFAIK.



  • Darn.  I'm thinking about switching back to pfSense but I really want fq_codel.



  • fq_codel, the ZFS of AQMs, or nearly. Cake aims to be the "ZFS", but close enough.



  • Is the command of "ipfw sched 1 config pipe 1 type fq_codel && ipfw sched 2 config pipe 2 type fq_codel" the same if I only have 2 root limiters?  Both of them are root limiters one has a mask of source and the other has a mask of destination.

    I would like to try this out but wondering if the command is different for just root limiters without "child" queues.  Obviously I am highly dependent on the gui I am a bit confused with the ipfw command since it references both sched and pipe.

    Thanks for any reply!



  • TS sample is for the root limiters also, if you have  some troubles understanding, post the content of your /tmp/rules.limiter



  • here is the content of my /tmp/rules.limiter

    pipe 1 config  bw 100Mb mask dst-ip6 /128 dst-ip 0xffffffff

    pipe 2 config  bw 10Mb mask src-ip6 /128 src-ip 0xffffffff

    I need help with the ipfw command to enable fq_codel on pipes 1 and 2 because i don't have any child queues.

    thanks in advance



  • According to documentation posted in this thread you need to configure sheduler at least to make things work.

    
    pipe 1 config bw 100Mb mask dst-ip6 /128 dst-ip 0xffffffff
    sched 1 config pipe 1 type fq_codel 
    
    pipe 2 config bw 10Mb mask src-ip6 /128 src-ip 0xffffffff
    sched 2 config pipe 2 type fq_codel
    
    

    EDIT:
    Tested, it will not work. You need to configure child queues and use them in ruleset, exactly as described by TS. Default automatically created pipe queue always uses FIFO sheduler and I am not sure it is possible to change this.

    So after changes made in GUI also, you must edit and create your own rules.limiter that should look like this.

    
    pipe 1 config bw 100Mb 
    sched 1 config pipe 1 type fq_codel
    queue 1 config pipe 1 mask dst-ip6 /128 dst-ip 0xffffffff
    
    pipe 2 config bw 10Mb mask 
    sched 2 config pipe 2 type fq_codel
    queue 2 config pipe 2 mask src-ip6 /128 src-ip 0xffffffff
    
    

    So the right answer is no you can not shape with fq_codel using only root limiters.



  • Thats really too bad.  We use PFsense primarily to "specify bandwidth limits per host." for a small ISP.

    I really wish I could find a way to limit a subnet to say 100Mbs and then limit each ip host address in the subnet to 5 Mbs.  And then have each IP address dynamically shaped if the overall link was approaching the 100Mbs total.

    Is it possible to combine and use ALTQ and Dummynet at the same time?  Has anyone tried that or have a config example?

    I guess I could use limiters on 2 PFsense boxes.  First one limiting each host to 5 Mbps using limiters with a destination/source mask.  And the second limiting the entire subnet to 100Mbs using limiters without a mask and changing the type from WF2Q+ to FQ_Codel by issuing the command "ipfw pipe 1 config bw 100Mb type fq_codel"

    I hope thats not too confusing.  Anyone have a more eloquent way of trying this?

    As always, thank you for any reply.



  • Yes it's possible, but  you will have some overheads and losses, you can try it at least, I think. Just set your per host limits on ALTQ shaper side and do your evenly shared FQ_CODEL enabled limiters exactly as TS described for you entire subnet.
    I am sure it is possible to build ipfw only shaper model that works like you want it to work, but it would be complicated not only with pfSense and can cause some errors on pfSense.



  • Got this setup! Thank you so much! I have been waiting for a way to run FQ_Codel on my pfsense box for a while now. Granted it had to be hacked on but it worked!

    Has anyone been running Suricata with 2.4 and fq_codel? Until I removed the suricata package my connection would keep dropping and I had lots of issues. So far so good.

    I also had to enable Hardware checksum offloading and TCP Segmentation offloading. I may have to re-enable these at some point but at the moment everything is going well.

    My last speed test.



  • @cplmayo:

    Got this setup! Thank you so much! I have been waiting for a way to run FQ_Codel on my pfsense box for a while now. Granted it had to be hacked on but it worked!

    Has anyone been running Suricata with 2.4 and fq_codel? Until I removed the suricata package my connection would keep dropping and I had lots of issues. So far so good.

    I also had to enable Hardware checksum offloading and TCP Segmentation offloading. I may have to re-enable these at some point but at the moment everything is going well.

    My last speed test.

    enable re-enable or disable re-enable or enable re-disable or .. ?



  • @Harvy66:

    fq_codel, the ZFS of AQMs, or nearly. Cake aims to be the "ZFS", but close enough.

    This is very interesting.

    Any chance someone(s) knowledgeable would be willing to put together a single post along the lines of this - https://forum.pfsense.org/index.php?topic=126597.0

    Kind of like an fq_codel one-stop shop for the layman?



  • ok am finally testing this and got it working.

    I had observed some iptv/vpn issues that seemed to only occur when my ingress altq config was active, so am now testing this configuration.  I have not yet tested if this is as effective as hsfc alt for keeping steam downloads in check, I had to set the dummynet limiter to 95% of downstream cap to even get a 6 threaded downstream test to stop causing packetloss, so not confident that will be enough for a 30+ stream steam download but will see.

    How granular is this? can I e.g. route steam etc. all through it but at the same time applying a limit less than 95% for steam download whilst keeping things like youtube able to burst higher.  All on dummynet.  As I have a feeling I will need to drop this to at least 90% to manage steam but I consider that too low for lighter threaded stuff.



  • The percentage of bandwidth you pay for is situationally dependent. If you always get 100% of what your isp says they'll give you then 95% works. If it dips to 94% of what you subscribe for and you set dummynet to 95% then dummynet can't do anything for you.

    It's a granular as firewall rules can be



  • @belt9:

    The percentage of bandwidth you pay for is situationally dependent. If you always get 100% of what your isp says they'll give you then 95% works. If it dips to 94% of what you subscribe for and you set dummynet to 95% then dummynet can't do anything for you.

    It's a granular as firewall rules can be

    shaping is nowhere near that simple.

    The reason steam is harder to shape is it opens so many threads.

    I always get 100% from my isp but it doesnt mean 95% will always work well for all types of traffic.

    Looking at dummynet configuration it looks like multiple specific rate limits cannot be set within one pipe however weighting can be applied so I can put steam downloads on a low weight and things like dns lookups and emails on a high weight, this is what I will look into on my config next. Thanks to the OP giving me a starting point. :)

    AltQ on PFSense is incredibly granular but of course someone has put the effort into integrating it all into the GUI, and AltQ itself allows children in a queue to have their own limits set.

    Already got some good results.

    When I setup the dummynet config (basic as in the OP) I had a iptv stream running to my STB and I can see from my ping monitoring on my connection, the peak latency has plummeted, it was an almost steady increase in peak latency, now its spikes instead of constant and the spikes been generally much lower.

    I will post back on how my steam downloads testing goes.



  • That's kind of the point of fq_codel, KISS. It is by design intended to be simple.

    Now you might be trying to get it to do something it wasn't designed to do, in which case yes you will have to do some weird stuff - bit more likely you should just look elsewhere.



  • I am a fan of simple, if steam works well in the current config then the current config stays, the weighting is a fallback plan if it doesnt work well.

    From what I understand all the weighting is dummynet side, it simply dynamically adjusts the throughput allowance of each thread based on the weight assigned, by default everything has the same weight.

    All fully documented on the dummynet man page, I dont think its a non supported feature.



  • Yeah man page is very helpful.

    You should just be able to weight steam as you desire, apply the queue to a rule that catches ports and protocols for steam and let fq_codel do the rest.

    I think you'll be happy with it. I just set it up and it's working very well for me.

    It's awesome for weighting a guest net and primary net!



  • I played some more.

    First I misunderstood the man page, the weight flag does nothing on fq_codel, it only has an affect on another queuing type.

    I tested steam and the result wasnt good, lots of packet loss during a steam download, the packet loss only goes close to 0% when the pipe size is below 40% of my line capacity, as I said steam is probably the most brutal traffic I have seen on my home connection.

    HFSC can manage it at anything below 90%, however latency during saturation is vastly superior on fq_codel, packet loss is worse but latency better.

    If I increase the queue slots to 500 (default 50), then packet loss almost stops at pipe size below 75%, with a small hit to latency.



  • It sounds like something else is going wrong on your box.

    Weighting definitely applies to fq_codel, I've tested it on my own system and it matches weighted values every time.

    I've also tested with both steam and flent rrul. No packet loss.

    What is your line bandwidth? Are you trying to use dummynet and altq at the same time?



  • altq is off during this testing to ensure no conflict.

    Thanks for clarifying weighting has an affect I will try it on my test config as originally planned.  The reason I said it wasnt valid is because this is in the man page from dummynet section on ipfw.

    weight weight
                Specifies the weight to be used for flows matching this queue.
                The weight must be in the range 1..100, and defaults to 1.

    The following case-insensitive parameters can be configured for a
        scheduler:

    type {fifo | wf2q+ | rr | qfq}
                specifies the scheduling algorithm to use.
                fifo    is just a FIFO scheduler (which means that all packets
                        are stored in the same queue as they arrive to the
                        scheduler).  FIFO has O(1) per-packet time complexity,
                        with very low constants (estimate 60-80ns on a 2GHz
                        desktop machine) but gives no service guarantees.
                wf2q+  implements the WF2Q+ algorithm, which is a Weighted Fair
                        Queueing algorithm which permits flows to share bandwidth

    type {fifo | wf2q+ | rr | qfq}
                specifies the scheduling algorithm to use.
                fifo    is just a FIFO scheduler (which means that all packets
                        are stored in the same queue as they arrive to the
                        scheduler).  FIFO has O(1) per-packet time complexity,
                        with very low constants (estimate 60-80ns on a 2GHz
                        desktop machine) but gives no service guarantees.
                wf2q+  implements the WF2Q+ algorithm, which is a Weighted Fair
                        Queueing algorithm which permits flows to share bandwidth
                        according to their weights.  Note that weights are not
                        priorities; even a flow with a minuscule weight will
                        never starve.  WF2Q+ has O(log N) per-packet processing
                        cost, where N is the number of flows, and is the default
                        algorithm used by previous versions dummynet's queues.
                rr      implements the Deficit Round Robin algorithm, which has
                        O(1) processing costs (roughly, 100-150ns per packet) and
                        permits bandwidth allocation according to weights, but
                        with poor service guarantees.
                qfq    implements the QFQ algorithm, which is a very fast
                        variant of WF2Q+, with similar service guarantees and
                        O(1) processing costs (roughly, 200-250ns per packet).

    This made me think its exclusive to wf2q+ however, fq_codel is omitted on the type section, so it didnt confirm that fq_codel has no weighting algorithm so I made the assumption.

    My downstream throughput is around 71603kbit. Calculated after removing vdsl overheads, and also confirmed with experimentation when rate limiting to see when a rate limit starts having an affect.  The bandwidth is very consistent whether its on peak or off peak, if I remove shaping and download via steam it flatlines at the max speed with no dips.

    Steam on average opens between 20 and 40 connections when downloading, most of these connections appear to be short lived making them very difficult to shape. Instead of downloading a large compressed file and uncompressing it steam seems to either download individual files on their own sessions or download files in fragments with the aim of maximising tcp sessions.  The problem is significantly reduced if I choose a server that has a high rtt such as in america (I am in the UK), swamping a connection with low RTT high bandwidth sessions will murder it.

    I have been looking at the box configuration itself, the hardware tuning etc.  As I understand, if packets are batched together with things like interrupt moderation as well as a low kernel hertz timer, then shaping is less efficient as it cannot intervene at frequent enough intervals.  These are all things I am investigating as an ongoing process and I havent given up on this.

    I have just reported back here how things went on the configuration suggested in the OP, with the only alternative config tried so far been to increase the queue depth.

    So so far on my unit on my usage test patterns fq_codel via dummynet is much better for latency but worse at packet loss compared to HFSC on altq.

    Also to add steam itself supports throttling the speeds, in that case I have tested on "unlimited", "7MB sec, which seems to be just below what my line can do" and "5MB sec", when steam throttles its not clean tho, it works by spiking to full speed, then pausing, then full speed again so it evens out that way.  If I set my pipe size lower and leave steam at unlimited its a clean reduction in speed flatlined at the pipe throughput rate.  throttling via steam vs the pipe size is more effective at higher speeds, but the pipe gets better when set very low.  I will report back after trying more stuff and welcome suggestions that are reasonable (trying entirely new kit is not reasonable in case you about to suggest it).



  • You could try applying a different shaping algorithm to steam and see if it works better.



  • I have stopped playing for now and moved back to HFSC ingress and fairq egress.

    Suddenly thats improved due to the changes I made to try and help dummynet.

    I did some tuning based on the information here (the author of isr code)

    http://alter.org.ua/soft/fbsd/netisr/

    I disabled interrupt moderation on my 2 intel ports.

    I increased kernel timer rate which doesnt help interrupt overheads but I have the spare cpu clocks idle.

    Now ALTQ+ HFSC is handing steam no problem at 97%, the latency performance is much closer to fq_codel as well.

    I dont know yet if my iptv issues will improve or if I will still need to temporarily disable ingress ALTQ when relying on the iptv stuff.

    The netisr changes I consider experimental of course as it goes against what I considered good practice (and obviously what pfsense considers good practice) the author seems to suggest deffere mode for netisr is superior for routing and direct is only best when running as a server, especially the case when you have more cpu cores than network cards.

    Hopefully more people will contribute to this thread on their experience of dummynet fq_codel vs ALTQ shaping (particurly HFSC).



  • Increased the timer rate? It defaulted to 1000hz for me or at least I don't remember changing it. Was it lower for you or did you increase it even more? FreeBSD 10 added tickless kernel scheduling. I'm not sure kern.hz even applies anymore.

    I'm also getting good results from ALTQ+HFSC+Codel



  • yeah by default it runs double the kern hz on my unit, I had set it back to legacy behaviour so 1000khz timer, but for this I removed my override so its 2000 per second on each core.

    By the way I think now this is a PPS issue that my unit is failing to handle when its too high.

    It seems HFSC drops the PPS quite aggressively when it is in play and this is why it is working better for me, from what I can tell my unit starts dropping random packets when PPS is over about 1500 when it reaches 2500 or so it gets quite bad, although the download is never affected visibly.

    I do have NAT, no cpu cores are been saturated, pfsense is behind a vdsl modem I have in bridge mode so there is also a chance that modem is not handling the higher pps itself.

    So at least for now I will keep my feedback out of this thread given the problem for me might be related to me hitting a PPS bottleneck.



  • Sounds like something is wrong. My i5-3.2ghz Haswell Intel-i350T2 is handling 1.44Mpps with HFSC+Codel just fine. 400Kpps is about 15-20% load and 1.44Mpps is about 20%-25% load. And I still get about 1ms pings while doing this.

    I have some forum posts around about what changes I made, like enabling MSI-X (enable soft interrupts if your NIC properly supports it) and removing the limitation on the number of packets to process per interrupt, which is by default 40 I think.

    I used to process 1.44Mpps at only 600 interrupts per second, but over the years, it is now about 1,200 interrupts per second. All I know is CPU is low, interrupts are low, latency and jitter and loss are low. Even with only 300 interrupts per core per second.



  • agree something is wrong, problem is I dont know what at the moment. :)

    Will diagnose some more at some point.

    MSIX is enabled by default on my i350, I actually temporarily disabled it already to try and get to bottom of it but MSIX in this case is not the solution. The packets per interrupt is something I not heard off, you know where that is configured?


Log in to reply