Hi, whenever my RB3011 is under load I mainly see cpu0 hitting close to 99% load and cpu1 is idling at around 1 to 7%, is there a reason for this or do I have a misconfiguration? currently this is causing some throughput loss because whenever the cycles are being spent they are not shared between cores, it all happens on a single core...

Here's a quick snapshot of what it more or less looks like under WAN loads: (if the image doesn't show this is the url: https://imgur.com/a/CC6k4oV)

Any advice? I know I should probably upgrade but the fact that only one core seems to be under load is worrying me...
Thanks

Use Tools / Profile to have at least an idea which service is causing the load.

Some services can be distributed to multiple cores, some are single-core.
It can also be a config issue (e.g. wrong VLAN setup where everything passes CPU instead of being HW offloaded).

Thank you, I profiled for a while under maximum wan load and these were the biggest offenders:

firewall 17%
ethernet 5-10%
networking 10-12%

Interestingly for the firewall if I just set it to fast-track it still consumes the same amount of CPU cycles (according to the profiler).

This could also be an issue with the load balancing setup I currently use because one of the WANs is Starlink and it's faster than the fiber service I have so I tried some weighted balancing using PPC through Mangle rules. I went back to regular 2/0 and 2/1 with just two rules for an equivalent 50/50% which isn't great because I can't make full use of both connections unless there are many, many segmented connections (imagine a download manager with 30 or more fragments, at this point I get an overhead from each fragment as well).

I suspected CPU limitations because of what I saw showing just one core in use and not the other, also because separately I can get better throughput from each of those interfaces but not through PPC.

I know I should upgrade since it's an old router now, I bought it when it just came out and it has served me well, no complaints and I know V7 is far from ideal for this specific model -- Keep in mind I do use adlists and that's primarily why I upgraded to V7 (the previous method was atrocious in load times and memory use).

But since I'm no expert I figured asking those who are generally working with RouterOS / Mikrotik on a daily basis could have some insights on the matter.

I don't currently plan to have more than two WAN gateways but I'd like to properly balance them -- Should I get a separate gigabit load balancer? (there are some offers on tplink some used, some new... the used ones are 1U the new models are not rack mount -- go figure) the other option for me would be to upgrade to a modern 1U Mikrotik even though they are spendy but I may be able to start advancing toward a 10G setup in the future...

The other part of my setup involves a switch with just two LAG/aggregated interfaces but currently the tests have been through the AP which is tied directly to the RB. (I've also tried through ethernet directly on the AP's interface to a laptop but there's no throughput difference) I'm aware that one part of the problem could also be the Starlink router currently NOT setup in bypass mode, so there's two NATs at play here which from my understanding that's a bad idea in general and it may be adding to the problem.


Thanks again!

PD: no VLANs currently at play, I'm running bare-bones to figure out what's wrong first

You could have mentioned that load balancing setup to start with.

Rule of thumb to assess performance:
look at test results, routing and 25 filter rules.
That gives you an idea of the ballpark figure your device should be able to handle as a regular router.
For RB3011 it shows 453Mbps (give or take).
Probably tested using ROS6 given the age of this device so under ROS7 it might be a bit less.
But it gives you an idea.
(as a reference: for RB5009 that's 3096 Mbps. Quite a bit more)

You're not providing info on both WAN speeds nor what you are getting now using your current setup ?
Also, be aware that PCC setup, since it's using mangle rules, is fully CPU bound.
So that could very well be a next limiting factor.

Your setup might also be a limiting factor if the most used ports are not connected on the same switch chip (RB3011 has 2, see block diagram).
Communication between both switch chips has to pass CPU as well.

CPU load on itself is nothing to worry about unless one of the cores inches towards (or gets to) 100%, then you will be choking the pipe at which point you need to find where the bottle-neck is.
But it might very well be your options for improvement are rather limited with that device.

Thank you,

I was aware of the two switches, but I wasn't aware it would play such a big role in performance -- We learn something every day!
After I read your suggestions, I moved WAN2 to the first switch on the RB and removed the external switch with LACP for the time being as I test the WAN throughput on both sides.

Now both cores are being used and it seems to be splitting the load - Interesting.
Mind you I still have an interface enabled on the second switch of the RB but it's irrelevant in terms of throughput since it's just for IoT purposes.

Anyway, now it's stable (as it can be, given the nature of the second WAN) and I reached the expected speed, now I just need to figure out how to get the LACP setup for the external switch, maybe I can just use the remaining two interfaces from the first switch on the RB and that would be enough for most of the use at home... I'd like to have 4x though.

I'll try my luck "weighting" the load balancing setup once again, following this guideline: https://mum.mikrotik.com/presentations/US12/steve.pdf

This goes to show how important it is to stop, read and even draw on paper while keeping everything in mind... Since it's a home setup I wasn't too serious with the initial setup but that was clearly a mistake.

The current setup interface wise is:
WAN1
WAN2
AP
SW

You're not providing info on both WAN speeds nor what you are getting now using your current setup ?

Sorry, one is 100-100 symmetric and the other is variable in nature let's say 100-350 on the downstream and 10-20 on the upstream.


I still have a remaining interface for the switch to at least get a 2x LACP setup, but unsure on how to address it as of right now... If I move to the secondary switch block on the RB this would turn into the new bottleneck when it comes to WAN traffic I imagine...