I want to know how you can divide the traffic in different classes. So I created two classes, changes the weight parameter for different rates, generated some traffic and noted the bandwidth the two classes gets.
I used a second qdisc so the two classes are attached to a qdisc and not to another class.
Used script (download)
WEIGHT1=80 WEIGHT2=`expr 100 - $WEIGHT1` WEIGHT1="weight $WEIGHT1" echo $WEIGHT1 WEIGHT2="weight $WEIGHT2" echo $WEIGHT2 RATE_TOT=140kbps # In KiloBytes/Sec IP=kriek RATE1=112kbps PRIO1="prio 3" RATE2=28kbps PRIO2="prio 3" DEV="dev eth0" OPTION="allot 1514 maxburst 20 avpkt 1000" tc qdisc del $DEV root tc qdisc add $DEV root handle 10: cbq bandwidth 10mbit avpkt 1000 tc class add $DEV parent 10:0 classid 10:2 cbq bandwidth 10mbit rate $RATE_TOT $OPTION isolated bounded prio 3 tc qdisc add $DEV parent 10:2 handle 20: cbq bandwidth $RATE_TOT allot 1514 avpkt 1000 tc class add $DEV parent 20: classid 20:10 cbq bandwidth $RATE_TOT rate $RATE1 $WEIGHT1 $OPTION $PRIO1 tc class add $DEV parent 20: classid 20:20 cbq bandwidth $RATE_TOT rate $RATE2 $WEIGHT2 $OPTION $PRIO2 tc filter add $DEV parent 10: protocol ip prio 3 handle 1 fw classid 10:2 tc filter add $DEV parent 10: protocol ip prio 3 handle 2 fw classid 10:2 tc filter add $DEV parent 20: protocol ip prio 3 handle 1 fw classid 20:10 tc filter add $DEV parent 20: protocol ip prio 3 handle 2 fw classid 20:20 iptables -F iptables -X iptables -N acc_0 iptables -N acc_1 iptables -A OUTPUT -t mangle -p tcp --dport 2000 -j MARK --set-mark 1 iptables -A OUTPUT -t mangle -p tcp --dport 2001 -j MARK --set-mark 2 iptables -A OUTPUT -p tcp --dport 2000 -j acc_0 iptables -A OUTPUT -p tcp --dport 2001 -j acc_1
TESTA : RATE_TOT = 140kbps
RATE1 = RATE2 = 70 kbps
TESTB: RATE_TOT = 140kbps
RATE1 = 84 kbps (60%)
RATE2 = 56 kbps (40%)
TESTC: RATE_TOT = 140kbps
RATE2 = 112 kbps (80%)
RATE1 = 28 kbps (40%)
TESTD: RATE_TOT = 140kbps
RATE2 = 126 kbps (90%)
RATE1 = 14 kbps (10%)
TESTE: RATE_TOT = 140kbps
RATE2 = 133 kbps (95%)
RATE1 = 7 kbps (5%)
| WEIGTH1 | WEIGTH2 | TESTA | TESTB | TESTC | TESTD | TESTE |
|---|---|---|---|---|---|---|
| 50 | 50 | 50 % | 50 % | 50 % | ||
| 60 | 40 | 54.9% | 60 % | 60 % | ||
| 70 | 30 | 59.3% | 60.7% | 70.3% | 70% | |
| 75 | 25 | 61.2% | 62.7% | 75.4% | ||
| 80 | 20 | 62.8% | 63.9% | 79.6% | ||
| 85 | 15 | 64.2% | 64.5% | 79.9% | 85.6% | |
| 90 | 10 | 66.1% | 66.3% | 80 % | 89.2% | 90% |
| 95 | 5 | 69.3% | 69.5% | 81.3% | 90 % | 94.2% |
| 97 | 3 | 70.8% | 70.5% | 81.2% | 94.4% | |
| 98 | 2 | 73.4% | 72.9% | 82.5% | ||
| 99 | 1 | 78.2% | 78.1% | 85.8% | ||
| 100 (X) | 0 | 7.4% | 8.2% | 14.1% |
Giving one class 0 as weight is a bad idea. It disturbes the internal sharing and results in very strange results (TODO: test again).
I tried also to generate 10 streams in class 1 and 1 stream in class 1. But the results were the same (TODO).
At first sight, the results don't look good. But when you have the same proportion for the rates and the weights, the result looks good (red results). So as a rule, take the weight proportional the rate and everything will be fine. But when you choose the weight for one class a lot bigger, also the smaller proportionals in rates are very accurate.
When you remove qdisc 20: and attach the 2 subclasses to class 10:2, then is the weight parameter total ignored. The traffic is divided to the two sublcasses proportional to the rate of the two subclasses.
If you use a filter to put the traffic in 10:2 and then two filters to put the traffic in 10:10 and 10:20, each class get 50%.
I tried also to generate 10 streams in class 1 and 1 stream in class 2. But the results were the same (TODO: extra tests results).
I tested it with this script :
Used script (download)