Using the 1009 CCR and 1036 CCR routers, we're seeing very high CPU usage in processing simple queues when they use a parent. I'd like to say that I haven't seen this happen before in earlier versions of Router OS before the Bucket size was introduced, but in reading it looks like Bucket size was always used (just not editable). Maybe bucket size has nothing to do with it, not sure, looking for input.

Router 1 - 1009-7G-1C-1S+ with RouterOS 6.37.1:
Parent Queue manages 26 Child Queues with speeds ranging from 10mbps to 500mbps (half are 500mbps). If a lot of data is transferred or a speedtest is started, the CPU will increase to 50% plus for 500mbps speeds, and slightly lower on the lower speed packages. With some of our older setups, I don't recall seeing CPU rise so high when speedtesting is taking place.
If the child is removed from the parent, CPU rises minimally to 5% during a 500mbps speedtest.

Router 2 - 1036-8g-2s+ with RouterOS 6.37.3:
Parent Queue manages 280 Child Queues with speeds ranging from 100mbps to 1gbps (25% are 500mbps and up). Anytime a speedtest takes place on a connection with 500mbps and up speeds, CPU hits 100% and the speed tops out at about 300-500mbps. I removed all highspeed (500mbps+) queues from the Parent and CPU lowers drastically and speeds are near perfect when testing gigabit.

Testing on the 1036 with 280 Child Queues:
In Parent:
1000mbps: Uses 100% CPU on all cores and therefore it’s killing other people’s connections.
500mbps: Uses 100% CPU on all cores too.
100mbps: Uses only 10% CPU average. So 100mbps and lower works well.
Outside of Parent:
1000mbps: up to 30% CPU average over cores
500mbps: up to 25% CPU average over cores
100mbps: up to 10% CPU

*Keep in mind that there is roughly 10%+ usage already when performing these tests.

Are you sure you want to use Simple Queues?
Maybe Queue Tree is a more appropriate config for your purposes?

Maybe. I think the only issue with Queue Tree is that you lose statistics per connection, and it makes troubleshooting more difficult when dealing with customers.

When everything is dynamic that is true, but when the customers are more or less static you can
create the queue tree using a script and you get stats per child item when you make one for every customer.
(i.e. you do not rely on PCQ)

Advantage is that it performs much better and allows the use of FastTrack.

Interesting. I'd like to try that. Do you have an example script by chance?

Of course the details depend too much on your configuration.
There is a "Fasttrack friendly QoS script" on this forum that does similar things (and things you might like to add).

I have the same problem here using a CCR 1036 ros 6.45.6, to ensure link to business customers i use parent queue and tonight the CPU reached 100% in several moments and profiles saw that was the queue
with a 450 queues pppoe Is a parent to bandwidth guarantee business customers, and with that increases the ping of all this had never happened, just this week after I updated, will be a version bug?, anyone else with this problem?