Any users have a CCR in a production environment?

I've been extremely hesitant to deploy my CCR - but really need the fiber support at one of our sites.

I'm running OSPF w/ multiple areas, ECMP, PPPoE servers with ~600 users, Queue Trees etc...

All the basics seem to work on the bench, but just wondering if anyone has been brave enough to deploy these in a production environment.

We have deployed one as our core router for ~900 customers or so. There were some teething issues in the beginning but this is to be expected running on release candidates, currently the only real complaints we have are that bgp seems to be limited to a single core right now and our mangle rules for limiting netflix bandwidth consumption that worked very well on an old PowerRouter 732 had to be disabled because they would slowly ramp up in cpu usage until the ccr would lock up and crash or have to be manually power cycled. OSPF seems to be working just fine although I can't vouch for ECMP PPPoE or Queue Trees. Mikrotik needs to make some more optimizations so that RouterOS can actually take advantage of the hardware inside the ccr but as I understand it they are working on it. All in all if you're not doing anything to complex on it it's very stable and its blazing fast, before our mangle rules started to act up it was running non-stop and stable since 6.0 was released and since we disabled our mangle rules the only reboots have been intentional to upgrade firmware or software.

I have 2 CCR's with 36 cores, acting as MPLS/VPLS Aggrigiation nodes for PPPOE clients, unfortunately one of the routers keeps freezing and loosing all of its interfaces, which happens every 24 hours i solved it by removing some of the links to another RB1100.

Each router have 60 Vlans, 12 VPLS tunnels and approximetly 400 PPPOE Clients. Each router is routing an avarage of 600 Mbit/s. Simple Ques are pretty good on ROS6 so i gave up using que trees. No firewall no nat on my system.

ROS Version RC13.

Sorry for off-topic posting on this thread?

@rcwmoab

.....mangle rules for limiting netflix bandwidth...
How effective was these mangle rules and could i request a copy?

Sorry for the off topic reply

@n21roadie
They are very effective, they also cover youtube and other streaming services as well. It's been a while since I have dug into them and really examined them but if I'm not mistaken they work on a per video stream basis rather than a per IP basis so that 2 people on the same IP won't throttle each others streams. We run cambium equipment and cap each stream out at 8Mbps, this really helped with the frequent quality adjustments and re-buffering our users were experiencing on netflix due to the cambium platforms speed bursting. Unfortunately we bought them from someone we met at animal farm so I can't give you a copy of them but if you're interested I can dig up his info for you, I think we paid ~200$ for them.

Back on topic.
We run a DNS server on the CCR as well and it doesn't seem to be optimized for a multicore environment either. We have a dedicated DNS server and the CCR is advertised via DHCP as our secondary DNS but it will still frequently max out the core it's running on and become kind of slow. All in all it seems mikrotik still needs to optimize many parts of RouterOS for a multicore environment. Total CPU usage remains below 10% on average but I think this number is rather deceptive, if you check the per core usage one or two cores will be running 100% and the rest will be doing almost nothing. In spite of this the CCR is still an incredible router during peak usage we'll average ~200-300Mbps, ~60,000 open connections and ~80,000 total pps and the CCR pulls it off without a hitch, no additional latency, jitter, or packet loss. Granted this is only a fraction of the advertised routing pps it's still impressive considering that most of it is going through BGP and only running on one core. All in all the CCR is a great product especially for the price point, and given the proper configuration is absolutely production ready, the caveat being it doesn't seem to be production ready on all configurations like a true carrier class router should be. Sorry I can't give you more info for your particular configuration. Also please forgive me if some of my figures or assumptions are off or out right wrong I really don't understand how the tile architecture works and everything but our CCR decided to break all at once so I've only gotten 4-5 hours of sleep a night for the past week, makes the brain just a little bit foggy.

Thanks for the feedback.

I moved to Queue Trees and PCQ's awhile back because simple queues were eating up the CPU on our routers. To switch back to simple queues would require changing our customer management system to feed the correct radius attributes.. Not very difficult to do, but I'll look into that once I know 6.x is stable enough for all of our Router/PPPoE servers.

From your post I'm assuming Queue Trees are not working at all in 6.0/6.1... Is that correct?

EDIT:

To clarify Queue Trees aren't being used for anything but Address-List connection marking for customer bandwidth throttling. We're not shaping protocols.