PCQ queues changes after ROS 3.17

RouterOS General
1

Ive noticed two visual changes on ROS above 3.17 (3.19,3.20), but not seen on 3.17, when you have a setup with more than one PCQ leaf under some inner class; i.e.:

/queue tree
add burst-limit=0 burst-threshold=0 burst-time=0s disabled=no limit-at=3000000 max-limit=3000000 name=downstream packet-mark=“” parent=global-out priority=8 queue=default
add burst-limit=0 burst-threshold=0 burst-time=0s disabled=no limit-at=600000 max-limit=1200000 name=http packet-mark=http parent=downstream priority=2 queue=pcq_dn_http
add burst-limit=0 burst-threshold=0 burst-time=0s disabled=no limit-at=2000000 max-limit=3000000 name=http_burst packet-mark=http_burst parent=downstream priority=3 queue=pcq_dn_http_burst

/queue type
set default kind=pfifo name=default pfifo-limit=30
add kind=pcq name=pcq_dn_http pcq-classifier=dst-address pcq-limit=30 pcq-rate=600000 pcq-total-limit=2000
add kind=pcq name=pcq_dn_http_burst pcq-classifier=dst-address pcq-limit=30 pcq-rate=2000000 pcq-total-limit=2000

that 1) when you use “wlan1” instead “global-out” on the inner class, pcq queues creates ONLY ONE subqueue per leaf (pppoe server on wlan1), no matter how many hosts actively using http you have on wlan1 atm; 2) the inner class does not show the correct average data rate, as you can see up to 2mbits on “http_burst”, but about 300kbits on the inner class average bit rate only.

Anyone else can confirm this? Please.
Thanks
Ozelo

2

Could this be a similar problem? Are you aware of the changes in the new versions? http://forum.mikrotik.com/t/queue-tree-default-queue-shows-bad-queued-bytes-value/26225/2

3

yes, ive tested the lastest ros version and the queue still the same… only 3.17 seems to show the correct avg rate and none creates more than one subqueue when attached to the physical interface.

4

Global-out is before src-nat, interface queue is after the src-nat. so from packet flow point of view it is logical. Are you sure that before 3.17 it was working???

This sounds like a bug

5

Global-out is before src-nat, interface queue is after the src-nat. so from packet flow point of view it is logical. Are you sure that before 3.17 it was working???

No, this was not always like this. I’m not sure when, but on earlier versions pppoe packets (L2) couldn’t be handled on interface queue. I mean, the same setup (mangle on postrouting, queue tree on interface queue) and then, before 3.17 but not before 3.13, pppoe packets could be handled at interface queue. This was a surprise for us, then we starting moving from global-out to interface queue. The increase on customer complaints lead us to investigate and in this case, turning on the column that show “PCQ queues” shows only 1 subqueue per leaf no matter how many users you have there. Right at this moment, moving from interface queue to global-out you can see more than only one. I’m pretty sure that this is a side effect of changes made before 3.13…

What’s new in 3.12:

*) fixed problem - static queues on PPPoE, PPTP, L2TP interfaces became invalid
on client reconnect;
*) changed behaviour of simple queues - queues with no limit and type default-small
and no children actually do not get installed, as if there was no queue;
*) graphing - fixed crash when dynamic interfaces/queues disappear;
*) improved PCQ queueing algorithm;

Kinda lot of changes on queues, isn’t? So on, we noticed that starting on 3.18 PCQ queues (as a leaf of course) does not show the correct AVG RATE on their “inner class” (not a PCQ ones), that should be the sum of all its children. I agree that sounds like a bug, but we have no complaints yet. Then I think it is a “visual bug”, not a damned one.

I was so curious if anyone else notice that, well…
Thanks
Ozelo

6

Since there is no more replies, thought I could post some evidences:

Picture 1: Illustrate how the inner class does not correctly sum leaf averages.

Picture 2: Illustrate how the leaf classes does not create more than one subqueue when the inner class is attached to the physical interface, as also don’t sum as picture 1.

Both pictures are taken on ROS 3.20… This is happening since 3.18
Thanks
Ozelo