We’re running 3.5K subs across 6 OLTs, currently our setup is like this:
ISP1, ISP2 —> CCR1072 (NAT, PCC) ----> BEQUANT (TCPO, SHAPING) —> CCR2116 (NO NAT, NO QUEUES, ONLY PPPoE) —> CRS326 —> OLTs (1-6)
I’ve setup Fq-codel in a queue tree (CCR2116) in the OLT’s vlans like this:
/queue type
add kind=fq-codel name=fq-codel
/queue tree
add max-limit=1G name=OLT-1_DOWN packet-mark=no-mark parent=vlan200_INTERNET_OLT_1 \
queue=fq-codel
add max-limit=10G name=WAN_UP packet-mark=no-mark parent=\
bonding1_BEQUANT queue=fq-codel
add max-limit=1G name=DOWN2 packet-mark=no-mark parent=vlan203_INTERNET_OLT_2 \
queue=fq-codel
add max-limit=1G name=DOWN3 packet-mark=no-mark parent=vlan205_INTERNET_OLT_3 \
queue=fq-codel
add max-limit=1G name=DOWN4 packet-mark=no-mark parent=vlan207_INTERNET_OLT_4 \
queue=fq-codel
add max-limit=1G name=DOWN5 packet-mark=no-mark parent=vlan209_INTERNET_OLT_5 \
queue=fq-codel
add max-limit=1G name=DOWN6 packet-mark=no-mark parent=vlan211_INTERNET_OLT_6 \
queue=fq-codel
We’re doing 6Gbps at peak.
I see sometimes a single core can reach up to 80% usage, average usage 19% like normal. I think it’s ok, not even 100%
Now, my question is, is this a good fq-codel implementation? Is it even doing anything (good or bad lol). I don’t want to shape clients in the CCR because Bequant is doing that, but I think they don’t use any SQM at all.