CCR1036 PPPoE 1000+ clients 400+ Mbit/s degraded perfomance

RouterOS General
1

Hi, we are using a CCR1036 RouterOS 6.6 as a PPPoE concentrator with simple queues for shaping. In the evening the number of online PPPoE connections reaches 1500. I started to notice that in the evening when the number of online PPPoE sessions exceeds 1000, clients’ throughput starts to degrade. Total throughput stays at 400-500 Mbit/s. CPU load does not exceed 20-30% but the CCR cannot go beyond 400-500 mbit/s.

Today I upgraded to 6.9 but I doubt it will make any difference. Has anyone experienced the same problem? What can be done apart from installing a second PPPoE concentrator and load balancing between them.

2

How did you check the CPU load? Is it the average load (from /system resource), or max load per core (from /system resource cpu)?
When the avg load is 30% - all cores can have 30%, but it’s also possible that 20 cores have 0% and other 10 have 100% - and that’s not good :slight_smile:. Always check all the CPU cores. If any of them have 100%, the performance can start degrading…

3

If I’m correct you could have much better PPPoE speeds with 6.9.
I believe MT changed from ppp to accel-ppp in RouterOS 6.8. It should be much faster.

4

Still capped at 400-500 Mbit/s, one of the cores is constantly at 100%:


ners@aad-rt-nas01] > /sys reso cpu pr interval=1

CPU LOAD IRQ DISK

0 cpu0 31% 31% 0%
1 cpu1 29% 29% 0%
2 cpu2 32% 32% 0%
3 cpu3 27% 27% 0%
4 cpu4 10% 10% 0%
5 cpu5 44% 43% 0%
6 cpu6 100% 100% 0%
7 cpu7 27% 27% 0%
8 cpu8 0% 0% 0%
9 cpu9 16% 13% 0%
10 cpu10 39% 24% 0%
11 cpu11 34% 33% 0%
12 cpu12 4% 4% 0%
13 cpu13 41% 41% 0%
14 cpu14 29% 29% 0%
15 cpu15 24% 24% 0%
16 cpu16 18% 18% 0%
17 cpu17 0% 0% 0%
18 cpu18 24% 24% 0%
19 cpu19 13% 13% 0%
20 cpu20 26% 26% 0%
21 cpu21 45% 45% 0%
22 cpu22 4% 4% 0%
23 cpu23 5% 5% 0%
24 cpu24 34% 34% 0%
25 cpu25 17% 17% 0%
26 cpu26 0% 0% 0%
27 cpu27 24% 24% 0%
28 cpu28 26% 14% 0%
29 cpu29 14% 13% 0%
30 cpu30 22% 22% 0%
31 cpu31 34% 34% 0%
32 cpu32 38% 38% 0%
33 cpu33 24% 24% 0%
34 cpu34 9% 5% 0%
35 cpu35 18% 18% 0%

5

Do you have dynamic simple queues or static? if dynamic, i would suggest to try out making them all static.
I noticed that removing and adding queues all the time results in performance degradation.

What other configuration do you have?
what does /tool profile menu says?

I suggest to give a mail to support@mikrotik.com

6

I observed the same behaviour with IPoE and PCQ queues - on 400-500 Mpbs clients begin to complain =( switched back to x86

seems like CCR is still not for shaping, while we’re successfully using them as simple firewall/policy routers at upto 2.5 Gbps and more (5 min average)

7

Thanks, I will have to load balance them then.

8

What is your opinion now ?

9

What other configuration are you guys running, i have ~700 customers on each of my boxes, though only around 140mbit at peak, dynamic queues, but my avg load is always at 0% on all cpu’s i mean once in a while one will spike for a half second to 10-20% but then back to 1-2%

I know you said the cap seems to be at 400mbit, but i just see something as odd that your load is so much higher than mine, you’d expect it to grow pretty consistently as bandwidth increases not exponentially. LOAD-IRQ-DISK all 0-1% on my box at 700 users…

Question when your not at that cap of 400+mbit, what does your CPU look like earlier in the day?

Also did you contact mikro support to see what they say?

Note i am on 6.24 as their were many CCR changes made between 6.9 and 6.24

10

no changes, we didn’t test latest versions in CCR shaping, still using x86 :slight_smile:

11

Are you running just one PPPoE Server/Port for your client authentication/traffic ?