Hi.

I have a CCR1009-8G-1S-1S+ with RouterOs veresion 6.15, about ~1700 simple queues and ~300 firewall rules. Max CPU usage 70%. I cant reach queue max limit. I can only achive about 70% of speed. For testing purposes I left about 400 queues, then speed limitation works correctly. Is MikroTik/CCR have problem to handle so many queues? Have anybody elese use CCR for bandwidth control in production enviroment with thousands of rules/queues ??

Do you have dynamic change mss in mangle? (one per Connection) ?
How about using pcq?

I dont change mss. This CCR is not a pppoe server. After reading this document http://mum.mikrotik.com/presentations/RU14/megis.pdf , I set simple queues because routeros support multi-core processing for those queues.

I use PCQ queues, does these work on multi core systems?
RB1100ahx2 or CCR ?

Hi.

I have a CCR1009-8G-1S-1S+ with RouterOs veresion 6.15, about ~1700 simple queues and ~300 firewall rules. Max CPU usage 70%. I cant reach queue max limit. I can only achive about 70% of speed. For testing purposes I left about 400 queues, then speed limitation works correctly. Is MikroTik/CCR have problem to handle so many queues? Have anybody elese use CCR for bandwidth control in production enviroment with thousands of rules/queues ??

What kind of throughput are you getting?

Same problem here. We have 80 dynamic simple queues (one for each PPPoE client coming from radius server) and 30 firewall rules. When the pppoe client makes a speedtest at speedtest.net or other he only gets with an 16 Mbit queue maximum of 12 mbit.

queue types are standard dedault small -> pfifo limit 10.

I use PCQ queues, does these work on multi core systems?
RB1100ahx2 or CCR ?

We are considering going down that path, and as a test put about 500 customers on a 1036 CCR. CPU maxes at like 14% (ram: 15GiB free), but we can't get our bandwidth through. The graph I attached shows maxing out around 580Mbps. It should be at least 200Mbps more than that. It's possible I have the ports/bridges configured in a non-optimized fashion. I typically use bridges as the glue across ports; especially on the CCRs which don't have an internal switch. The same router has about 40 firewall rules and a few mangle rules for routing. We use simple queues with PCQ to manage each account:

The upstream router is a 1009 CCR being used as our edge router (non-BGP with a 10Gbps fiber and service on it). It averages around 50% CPU and we're already in the process of upgrading it. Plugged directly into this router, I can push our download to 1.8Gbps pretty easily.

We could probably get a beefy server (4Ghz CPU, 8 core or something) and use that as our queuing server, but that seems like it's solving the symptom of a bug, not a matter of solving a problem of scale. We're already looking at other platforms to provide queuing as we're starting to get complaints by our customers.

Can we get any official input on this? :S