    Thanks for the tips Brian,

    Actually, I have many sorts of links, line PPOE ADSL, PPPOA ADSL which a use PPTP over PPOA relay, radio links that connect in the ethernet interface and cable modems.
    To change the SFQ queue size I must recompile de kernel? I think a saw some messages talking about that.
    One other thing, I have the 1:10 a 1:20 class, let's assume there is no voip traffic and all bandwidth is being consumed and it is in the other class. When the voip traffic starts, there is a inicial delay untill the 1:20 class starts to free bandwidth to the 1:10 class, as I've noticed. Should I change the burst and cburst parameters to get a better response or just make de queues smaller?
      Thanks.

Brian Carrig wrote:

On 16 Jul 2004 at 13:35, Alessandro Ren wrote:

 Hello Brian,

 This is the basis for the wshaper.
 I have only two classes and I put voip on the 1:10 class and the rest in the 1:20. I

am not

listing here, but I have the rule marking packets and sorting then into classes,

actually, I just put

one port into the 1:10 class, that's the voip port and nothing else. I really want to

keep the best

quality I can for voip, without bandwidth waste., because, if a page takes 1

seconds longer do

load is ok, but if a voip packet starts to get delay, we got a problem,
 I think, I must have no queue for voip packets, all packtes should be forwarded as

soon as

they get to the box, right?


You do actually have a queue for VoIP, as you implement SFQ for both the 1:10 and 
1:20 classes. To the best of my knowledge the default setting for this queue is 128 
packets. This may be too large for VoIP if latency is a concern so I would suggest 
making this queue much smaller (limit option). Unfortunately without knowing the 
particulars of your link I am unable to suggest a figure but have play around and see 
what suits.

Regards
Brian

#
tc qdisc add dev $DEV root handle 1: htb default 20

# shape everything at $UPLINK speed - this prevents huge queues in your
# DSL modem which destroy latency:

tc class add dev $DEV parent 1: classid 1:1 htb rate ${UPLINK}kbit burst 6k

# high prio class 1:10:

tc class add dev $DEV parent 1:1 classid 1:10 htb rate $[4*$UPLINK/10]kbit \
 burst 6k prio 1

# bulk & default class 1:20 - gets slightly less traffic,
# and a lower priority:

tc class add dev $DEV parent 1:1 classid 1:20 htb rate $[6*$UPLINK/10]kbit \
 ceil $[10*$UPLINK/10]kbit burst 6k prio 2

# all get Stochastic Fairness:
tc qdisc add dev $DEV parent 1:10 handle 10: sfq perturb 10
tc qdisc add dev $DEV parent 1:20 handle 20: sfq perturb 10


Brian Carrig wrote:
    We run something similar allowing customers to place traffic into gold, silver or 
    bronze classes. I reserve a fixed amount of bandwidth for each class and allow

them

    to borrow (I hate the idea of bandwidth going to waste). However, excess

bandwidth

    is offered preferrentially to gold, then silver before being offered to bronze.
    Because p2p and other bw consuming traffic are unlikely to be in the gold and

silver

    classes (they cost more) there haven't been any problems.
    I haven't really looked at the wondershaper script in much detail, how is voip

traffic

    prioritised?
    
    Regards
    Brian
    
    On 16 Jul 2004 at 12:19, Alessandro Ren wrote:
    
      
    
     I've been using a altered version of the wshaper script to priorize voip traffic for 
        
    my 
      
    customers.
     I'd like to know if someone in the list has any tips on QoS for voip, if someone 
        
    has done some 
      
    experimentation.
     I am using HTB and if someone on the LAN uses a p2p program, I started to 
        
    noticed in the 
      
    voip, with cuts, jitter and lag. If a reserve a fixed amount of bandwitdh not letting 
        
    anyonbe 
      
    borrow, it works fine, but then if noone is using voip, I have bandwidth going to 
        
    waste.
      
     I think I need some fine tunning oin the HTB parameters, but I am not sure sure 
        
    about that.