1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

NT-R16, 80Mbps QoS chokes on all but stock Tomato USB

Discussion in 'Tomato Firmware' started by Arch Duke of Awesome, Feb 4, 2013.

  1. Arch Duke of Awesome

    Arch Duke of Awesome Serious Server Member

    Hello gents, welcome your insight. (edit: title has a typo.. should read RT-N16)

    I have an ASUS RT-N16 rev1 running stock Tomato USB (Tomato Firmware v1.28.9054 MIPSR2-beta K26 USB vpn3.6) which runs beautifully with QoS enabled with my 80Mbps downstream. Typically I see about 40-60% CPU utilization on the sirq mostly. If I flip to any derivation of Tomato USB, toastman, shibby, etc... and max my downstream (not torrents, but usenet/nzb) my sirq hits 100% and 1/5/15 goes through the roof...

    The situation improves on the mod based firmware, if I reduce my throughput to about 40Mbps. It appears only the stock Tomato USB is capable of managing QoS at 80Mbps.

    Any ideas on how I could resolve this, or is it simply a limit I cannot avoid?

    Thanks in advance.
     
    cloneman likes this.
  2. cloneman

    cloneman Networkin' Nut Member

    +1 this.

    I have seen this happen as well. Perhaps its because Toastman has the special incoming QoS rules that actually work (with the global max inbound bandwidth), which causes a lot more traffic to be processed by QoS.

    @ 15mbps on toastman, (1 single http transfer), I see 40% CPU usage w QoS on, which means that probably I cannot do more than 40mbps. Some of my friends have 50mbps and 75mbps and they cannot turn on QoS on toastman or Shibby without their CPU peaking 100%.

    It would be nice to have some workarounds for this (maybe a way to turn off QoS downstream for customers with huge downstream pipes?), or just know what's causing it. (i'm not using L7 or ipp2p rules)

    What's nice though, is if you set it to max inbound for 40, the router is actually able to throttle traffic correctly.
     
    Arch Duke of Awesome likes this.
  3. Arch Duke of Awesome

    Arch Duke of Awesome Serious Server Member

    Thanks Cloneman, you comments validate my experience. My sense is that (like you mentioned) Toastman, Shibby, etc inject some enhancements into the base QoS code that cause an overhead on the max traffic/bandwidth they're able to effectively manage (as you said, above 40ish). Like you also, I tried adjusting/simplifying my rules, but whatever is spiking the CPU it does not appear to be related to the rule complexity.

    I really appreciate your quick response and validation, I wonder if others have found any work-arounds on what appears to be just a limitation. Cheers.
     
  4. Porter

    Porter LI Guru Member

    Sorry guys that QoS is demanding so much CPU power. Can't really help you there, if you already disabled the L7-filters. With my small 4,5MBit downlink even a WRT54GL doesn't really struggle...
     
    Arch Duke of Awesome likes this.
  5. Arch Duke of Awesome

    Arch Duke of Awesome Serious Server Member

    Seems so Porter, and yes - its difficult to get much sympathy on 80Mbps fibreoptic - but just I would much rather run toastman, etc than stock Tomato USB. Definitely a 'first world' problem.... but was hoping someone out there might know of a work around :)
     
  6. Toastman

    Toastman Super Moderator Staff Member Member

    Without having access to these speeds we can't experiment. But the workaround would seem to be to turn off QOS, or accept the fact that today's speeds are exceeding the capabilities of cheap low power home routers with such puny processors. Sooner or later some manufacturer will wake up to this fact and market one, I'm sure, but at the moment they are all getting excited about speed increases that are not exactly "giant steps for mankind" :p
     
  7. cloneman

    cloneman Networkin' Nut Member

    If you really wanted to, couldn't you setup a fake ISP on your LAN? Dual nat and such
     
  8. Arch Duke of Awesome

    Arch Duke of Awesome Serious Server Member

    Thanks Toastman, in a world where my phone is dual (some, quad) core, I would agree router manufacturers haven't keep pace at all. Since the stock Tomato USB does work just fine at 80Mbps, I'll just stick with it for now (as I do enjoy QoS)

    Appreciate your confirmation on the issue and cheers on your efforts in general sir!
     
  9. pharma

    pharma Network Guru Member

    We do miss TeddyBear. He performed a couple networking tweaks with Tomato USB that have not been duplicated.

    However it is good to know if I require 80mbps networking speeds which firmware to use. :)
     
  10. phuque99

    phuque99 LI Guru Member

    I can confirm that TomatoUSB can exceed 150Mbps with QoS enabled on E3000. Perhaps you were limited by RT-N16 hardware.
     
  11. Arch Duke of Awesome

    Arch Duke of Awesome Serious Server Member

    Stock Tomato USB? (i.e. I can run just fine with 'stock' (see my exact version, above) is why I ask as the E3000 runs at the same CPU clock speed mate..)
     
  12. eahm

    eahm LI Guru Member

    Shouldn't 600 MHz of the N66U be already enough for 150mbps? Here in Phoenix they're starting offering 150mbps cable and I wouldn't like to give up Tomato.
     
  13. phuque99

    phuque99 LI Guru Member

    I'm not sure what "stock" means to you. I compiled my own slightly modified version of Teddy's "most recent" TomatoUSB, from about two years ago. The modification are removal of some modules that I don't use and addition of some binaries that I want. Netfilter is unmodified, so it's the same code that handles the QoS and routing.
     
  14. Toastman

    Toastman Super Moderator Staff Member Member

    Remember that version would have the old stock QOS in it too, which would be quite possibly faster than in my build, because it now has proper ingress QOS. Any extra code or feature that is running will of course consume resources and reduce the maximum throughput.

    These consumer routers are not really designed for the speeds now starting to appear in some countries, people need to think about buying something with processing power that is up to the job without using smoke and mirrors to make the specs look better on the box ...
     
  15. pharma

    pharma Network Guru Member

    Toastman .... If I disabled QOS in your version should I be getting the faster speeds similar to TomatoUSB(with or without QOS enabled)?
     
  16. Arch Duke of Awesome

    Arch Duke of Awesome Serious Server Member

    That explains why phuque99 and I are fine with QoS enabled at higher speeds - so would it be fair to say the 'proper ingress' done right introduces additional demands on the CPU (and really 480-600Mhz is weak to begin with)

    Regrettably to your earlier point Toastman, I cannot find a better chip that supports Tomato....
     
  17. Engineer

    Engineer Network Guru Member

  18. Arch Duke of Awesome

    Arch Duke of Awesome Serious Server Member

  19. Pentangle

    Pentangle Serious Server Member

    First post here, but i've been deploying Tomato routers for customers (and internally) for several months now - thanks for the fine work guys!

    Reason I registered and am about to pipe up here, is that I wonder if this is a kernel issue, since it's similar to something seen in OpenWRT and other things, and I wondered if this might point towards an answer: http://lists.willowgarage.com/pipermail/pr2-users/2011-January/001367.html

    Incidentally, i'm sure Toastman will agree, but if your outbound bandwidth doesn't get saturated and/or you're not using amazingly bursty protocols and expecting them to co-exist with VoIP then that's 90% of the battle won with respect to the need for QoS on your link. That said, the easiest way to see the value of QoS is to marry a woman who spends half her day streaming films over the same 3mbit/s line you're trying to work on :)
     

Share This Page