I have an openbsd box doing queuing for 200+ users, each with their own cbq
queue to limit bandwidth on a per-client basis. My issue is that I'm seeing
a good 60-80% of the traffic on the client-facing interface going into the
default queue, rather than it going into the individual client queues.
my queuing is set up thusly:
altq on $cus cbq bandwidth 100Mb queue { cusdefqrx, resqrx, torrqrx }
queue cusdefqrx bandwidth 10% priority 1 cbq(borrow,ecn,default)
queue torrqrx bandwidth 1% priority 0 cbq(ecn)
queue resqrx bandwidth 89% priority 5 cbq(borrow,ecn) { 0000493qrx,
0000026qrx, 0000025qrx,
0000024qrx, 0000023qrx, 0000502qrx, ....etc }
and the client queues:
queue 0000493qrx bandwidth 1013Kb priority 1
queue 0000026qrx bandwidth 5940Kb priority 6
queue 0000025qrx bandwidth 2475Kb priority 1
queue 0000024qrx bandwidth 2475Kb priority 1
queue 0000023qrx bandwidth 1013Kb priority 1
queue 0000502qrx bandwidth 1013Kb priority 1
...etc
and then there are a bunch of rules that assign traffic to each queue:
x0000493="{ 192.168.4.136/32 }"
pass out quick on $cus from any to $x0000493 label 0000493rx queue
0000493qrx no state
pass in quick on $cus from $x0000493 to any label 0000493tx no
state
...etc, for every IP address in use, such that all traffic on the interface
is assigned to a queue.
pftop -s 1 -v queue shows:
root_fxp1 100M cbq 0 731630 771766K 0 0 0 0
0 893 952K
cusdefqrx 10M cbq 426563 519181K 0 0 0 0
0 513 737K
torrqrx 1000K cbq 0 0 0 0 0 0 0
0 0 0
resqrx 89M cbq 5 0 0 0 0 0 0
0 0 0
The problem is that queuing for each client doesn't seem to be working.
tcpdump shows traffic to a particular client which is not being assigned to
their queue. What really odd is that the labels for that rule ARE working,
so that traffic is hitting that rule.
Is there a way to see what packets are going into the default queue?
thanks for any insight.
David
