Hotspot performance / ROS does not use all CPU-power available

RouterOS General
1

Hello

I have a hotspot on a CCR1016. The hotspot is active on a bridge, connected to access points via EoIP tunnels.
Bandwith on site is approx 80Mbps/20Mbps. But i have never reached more than 35-40Mbps in real conditions. When there are many clients (150-200 active hotspot users), i notice that CPU ressources are badly spread among the 16 CPUs.
1 CPU is at 80-95%, and 15 others are relatively sleeping, around 5-10%. So I think this is the bottleneck.

I have made a screen capture with lower traffic, but we can already see the problem: only 1 CPU really working.
hotspot.png
I’m not trolling Mikrotik with this post (CCR1016 capped at 40Mbps rofl lol lol rofl etc); my current conclusion is that CCR with plenty of cores is not the device I need.

Does the RB1100 have much more power (per core) than CCR series ? What kind of Mikrotik gear do i need here ? Or do i have to buy a rack PC server(not the same price), put ROS on it ? Or is there any way of tuning CPU ressource allocation on Router OS?

2

I highly doubt being a problem of CPU, CCR1009 for example can cope with such load without problem.

I suspect the problem could be related to the EoIP tunnels, I’d try current version, EoIP / fastpath has been worked on since 6.33.

Are those tunnels using IPSec?

Is the firmware up to date?

3

I have activated fast path on all EoIP tunnels, will see if things get better.
Firmware is up to date (3.27) on routeros 6.34, and there is no IPsec encryption.

4

Are the other end of the EoIP tunnels mikrotik devices? A better performing option would be using MPLS/VPLS between them and your core.

5

Yes, Mikrotik on the other side. I’m not familiar with MLPS/VPLS, I’ll have a look at it.

6

Did fastpath alleviate the CCR load/bottleneck?

7

what kind of queues are you using??

some users have reported that situation on queues

8

@chechito:
I’m using simple queues (dynamically generated by hotspot). 147 active queues at the moment. In profile window, queuing takes less than 1%. The only values above 1% (approx 4-5%) are management and networking.

@pukkita
Fastpath did not give better results; 1 CPU@74% with only 25Mbps bandwidth. In the new screenshot below, you can see that only RX fastpath is used, TX fastpath is zero.. i’m not sure why
bva.png

9

Does that router have fasttrack enabled?

10

Nope, fasttrack interferes with simple queues, which i need.

11

Had to ask… :smiley:

Have you OSPF deployed on your network?

How are the APs connected to the CCR, directly by cable, PTPs (equipment)?

12

No OSPF.
APs are connected by LAN. We are using a bigger infrastructure so technically packets are using VLANs but the APs and controller are not aware of that.
If EoIP is the problem, i need to find something similar to transport layer 2 data over a layer 3 network, something less CPU-intensive (or better handled by routerOS).

13

That’s exactly what MPLS/VPLS provides, and more.

14

Oh damn i’m excited now, i’ll have to take a look at it this week-end :wink: