Re: [LARTC] Qeues seem to stop working after an hour, and it collapses, no idea why!

Linux Advanced Routing and Traffic Control

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Tuesday 01 April 2003 19:53, GoMi . wrote:
> Hello, this is my first message to the board, so please excuse me if i am
> omiting any prestablished rule ;)
>
> Let me explain you my problem..
>
> I have 300 users, studens, wich automatically means Edonkey,KaZaa,and most
> of this peer to peer sharing programs.
>
> I actually have 2 x 2Mbit ADSL's. My solution includes load balancing for
> those 2 inet access with a linux firewall separing both routers and my LAN.
>
> Right know i have a stateful firewall doing connection tracking (required
> for load balancing) DROPING all connections but web/smtp/pop3/https/imap
> etc.... That works fine, but people want to use p2p programs. After doing
> my research, i found out about QoS and ingress/egress queues.
>
> My idea was to priorize critical traffic like http/smtp/pop3/ssh/etc..,
> granting bandwith and leaving the rest to downloads (having these the
> posibility of ceiling bw as well).
>
> I have two egress and two ingress queues attached to the ethernets
> connected to each of the ethernet connected to the routers.
>
> Ingress is done with IMQ, and egress with HTB.
>
> At first it seems to work really well, but after half an hour having it
> working, INTERACTIVE TRAFFIC COLLAPSES, and looks as if there is no queues,
> hence interactive traffic is no interactive anymore.. :)
>
> I red somewhere that to decrease latency i had to set the txqueuelen from
> 100 to 30, but still collapses...
>
> I have noticed using iptraf, that the number of packets arriving to the
> LAN-eth (eth3 in my little sketch) is about 30% bigger than the sum of eth0
> and eth2, is that normal?
>
> I attach the configuration files of both the firewall, and the queues...
>
> And a little graphical description of my topology:
> Obviously doing SNAT and LB :)
>                   ______
> <hub1> <-------> |Switch|            ______
>
>                  |      |           |      |  eth0
>
> <hub2> <-------> |      |           |Linux | <----------> Router ADSL1
>   .              |      |    eth3   |      |192.168.3.5   192.168.3.6
>   .              |      |  <----->  | Box  |
>   .              |      |192.169.1.1|      |  eth2
>   .              |      |           |      | <----------> Router ADSL2
> <hub24> <------> |______|           |______|192.168.4.2   192.168.4.1
>
> I am really desperate, because it has been a month right now, and i am
> begging to suspect i am a bit of a fool ;) Thank you very much!!
I don't know the solution to you problem.  But I  have some remarks.

The burst you configured, is too low.  If you remove the burst parameter, htb 
will calculate the lowest prossible for you.
Also, just as test, give each class the same prio.
Pref (or prio) if you add a filter is only used to determine the order the 
filters are checked.  In your case, the pref doesn't mather so you take the 
same pref for the filters.

And is it each time after the same second/minutes that you have problems with 
interactive traffic?  And/or does it depends on the load of the link?


Stef

-- 

stef.coene@xxxxxxxxx
 "Using Linux as bandwidth manager"
     http://www.docum.org/
     #lartc @ irc.oftc.net



[Index of Archives]     [LARTC Home Page]     [Netfilter]     [Netfilter Development]     [Network Development]     [Bugtraq]     [GCC Help]     [Yosemite News]     [Linux Kernel]     [Fedora Users]
  Powered by Linux