In a previous thread I discussed my challenges with dynamic simple queues: http://forum.mikrotik.com/t/ccr-1036-pppoe-server/62520/8
As I wrote, when I changed from queue type default-small to default, performance rose and it appeared as my setup with PPPoE over EOIP and dynamic simple queues did not slow things down.
However, total traffic never exceeds 80-90 Mbps, even in peak periods.
This morning I switched to mangle and queue trees, lowering CPU from 5 to below 1%
Still awaiting the “verdict” when customers fire up their Netflix devices tonight…
With simple queues, at almost any time one of the CPU cores were maxing out, so I still suspect this to be the problem, i.e. that RouterOS does not distribute the queuing tasks to several cores, thus slowing total BW down.
With PCQ and queue trees no core exceeds about 10%
Googling the problem with simple queue performance on multicore HW, I came across the follwing ideas:
1) Change to MQ PFIFO
However, this is not allowed for other than ethernet interfaces, if I got it right.
2) Create the simple queues as sub-queues under queue parents
This is said to increase performance. I assume this is due to the queuing tasks of each parent being distributed to different cores.
Could I just create a number of parent queues and then specify these in property “Parent Queue” of the respective PPP profiles?
Then what would be the target of these parent queues?
The dynamic queues for PPP is created for each the particular PPPoE interface, so how would I manage to run them under parents?
I assume I’ll be happy with queue trees now, but I’m curious what actually slowed me down with simple queues.
It looks like the implementation for simple queues does not really utilize multi-core routers.
Can you shed some light on this Normis?
Or anyone?
