QOS:Combining Crawl/P2P/FilexFer classes

Discussion in 'Tomato Firmware' started by fefrie, Dec 1, 2013.

  1. fefrie

    fefrie Networkin' Nut Member

    I have a 7.5m DL Cable connection and I have each of the above imcoming classes set at a max/min of 6m/0m since I don't want any of those classes saturating the connection.

    I have the www class set at 6m-25m

    I have a cable connection that has burst speeds of up to 25m, so I've set the incoming limit fairly high at 25m.

    I realize that when I set the incoming limit at 25m there really isn't any QOS happening if all three classes aren't each maxing at 6m, and in fact could theoretically clog higher classes that need download bandwidth since theoreticaly, at 6m each adds up to 18m which far exceeds the actual download speed, but not the set incoming at 25m so there is no QOS working properly.

    This is a compromise I guess that is needed to get both high burst speed, and sufficient throughput for downloading (crawl/p2p) and FileXfer (Netflix streaming) with some sort of reasonable surfing experience.

    In reality, the three classes combined don't always saturate the connection to near max and I do get burst speeds for www (youtube). So it sort of works.

    The real QOS that happens is with the Outbound settings. According to the Toastman thread, Inbound and outbound QOS need to work together, but proper outbound settings are more critical than inbound settings.

    I only have a 0.5m upload connection, so with that I have www @30-100% and C/P2P/fXf each set at 1-70%

    Since outbound is so limited, and I've set the rules for outbound properly, it all seems to work.

    Inbound might be clogged with lower class traffic, but when I want something from the WWW, since the www outbound class has real priority, inbound www gets priority by association. There might be some initial inbound www packets lost, but as outbound www takes priority, all other outbound classes get reduced and their inbound packets get reduced too, making way for inbound www.

    It's all about the acks.

    Not sure if there is a question here, but if anyone wants to pick apart my thinking here and give me more insight, I'd appreciate it.

    QOS is hardly a science and it's definitley different for each user's experience.
  2. Marcel Tunks

    Marcel Tunks Networkin' Nut Member

    Queuing theory is math, but the way Toastman constantly tests and reevaluates his QoS implementation is definitely a science. :)

    QoS is hard to implement when the connection speed or quality are variable. You'll need to control outbound traffic primarily in your case. More recent kernels have tools built in that can help in such situations, but it appears that Broadcom doesn't care. :(
  1. This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
    By continuing to use this site, you are consenting to our use of cookies.
    Dismiss Notice