Or the whole MikroTik team was hacked 
lol this is a new low.. I’m getting trolled by a router vendor hahaha
+++1
- 1 Airtime Queue Limits (AQL) for IEEE 802.11
https://www.bufferbloat.net/projects/make-wifi-fast/wiki/:
This project focuses on reducing latency throughout the wifi stack, firmware, and hardware.
Ending the Anomaly: Achieving Low Latency and Airtime Fairness in WiFi:
https://www.usenix.org/conference/atc17/technical-sessions/presentation/hoilan-jorgesen
Seems likely, it coincides with IP>Cloud issues people were facing including me along with the forum timing out just about a week ago hahaha…
I don’t have any problems with SFQ on Tomato or AsusWRT, but I would like fq_codel on Mikrotik nonetheless, it sounds relatively easy to impliment. Please add it so I can move to a a more business grade platform and not lose features.
+1 for FQ_CoDel, +1000 for CAKE
For those that just need a replacement for SFQ, FQ_CoDel sounds great.
That’s Not Me!
I’m an ISP shaping traffic at the core. I need a Queuing Discipline (QDisc) that pays attention to DSCP. No QDisc currently available in ROS-6 pays any attention to any kind of priority markings on the packets. I have to build a HTB tree for each customer in the simple queues to be able to prioritize voice and video traffic over common traffic and demote bulk traffic. This results in 5 simple queue entries for each customer. But this only works for customers with Static IPs! I can find no way to build such trees for dynamic services who get their speeds from RADIUS. So all my PPPoE and RADIUS-DHCP customers suffer. If I had a QDisc like CAKE that DOES pay attention to DSCP, then the single simple queue that is dynamically created by RADIUS would work perfectly fine.
And one thing to check in your CAKE implementation. The original DSCP Code Point for LE traffic was CS1 (001000). According to https://tools.ietf.org/html/rfc8622, LE now has its own code point of ‘000001’
Finally an ISP that cares about providing low bufferbloat to its customers, kudos to you
Finally it is happening. Cake and Codel support in RouterOS: https://mikrotik.com/download/changelogs/development-release-tree
Thanks Mikrotik _/_
Nice!
Now all we need is to Mikrotik update their Queue Type documentation.
But one step at a time.. At least we’re moving in the right direction. Meanwhile I’ll play with it and see what breaks what.
Delivered on 7.1b3! Awesome, thanks Mikrotik.
Off to test!
I had problems enabling this in 7.1beta3 my CHR router would reboot every 20-30mins or so on Cake and a continuous reboot using FQ_CODEL
Tested on my CCR1016 with both Cake and FQ_Codel.
Cake performs poorly and limits my download a bit too hard.. perhaps it’s CPU loading.
FQ_Codel performs great and does the job.
EDIT: Further testing - switched from simple queue to queue tree (upload on wan and download on lan) and Cake performs well in that configuration.
My RB4011 reboots regularly using Cake:
jan/02/1970 00:02:55 system,error,critical router was rebooted without proper shutdown
jan/01/2002 01:00:00 system,error,critical router was rebooted without proper shutdown
jan/02/1970 00:02:55 system,error,critical router was rebooted without proper shutdown
jan/01/2002 01:00:00 system,error,critical router was rebooted without proper shutdown
So I upgraded to 7.1beta3 and went into the drop down queue type options… Did they name it something other than Cake and FQ-Codel?
No. They added Codel, FQ_Codel and Cake. However, you have to create a new queue type and assign one of those to that new queue type. It’s not defined by default. (it wasn’t on my install)
So, its possible to use cake or fq_codel to limit pppoe customers ?
Sorry, but no info because is too new.
Upgraded my Hex last night and removed my extremely large queue tree for a simple FQ_CoDel queue. So far so good, only anomaly was IP>Route not working correctly but I have more to read in that.
Thank you Mikrotik team for acknowledging your customers.
Question now is RB4011 or wait…
Steveocee,
That’s cool that fq_codel works on a Hex. Are you able to determine whether it uses more, or less CPU, than your large queue tree, or is that not a fair comparison?
Jeremy