Hi,

Just made few simple tests yesterday.

RB5009UG+S+IN / ROS 7.9 (with default conf)

My ITSP provider gives me an 2.5 Gbps internet connection which I was able to reach without any problem with my "old" unifi UDM SE.

by default the rb5009 operates @ 1.3 Gbps (speedtest / iperf3 tests)

cpu load was low and I noticed that CPU speed didn't go above 700 Mhz (350 Mhz mostly)
so I decided to fix CPU speed @1400 Mhz via winbox->system->routerbord->settings
Internet bandwith tests then increased painfully to 1.8-1.9 Gbps, but yes it was improved....

When I put the 2.5 Gb on LAN (bridged) and test bandwidth with and NAS (SFP+) then it is OK for 2.5 Gbps

Do I have to switch to a CCR2004 for ex in order to use my internet connecion at its top speed ?

I guessed that RB5009 could reach at least 3Gbps+ with 25 ip filter rules !

Thanks in advance!

I'm able to reach the 2.5Gbps speed using a GPON SFP and 3 PPPoE Clients with RB5009, So I don't think it is a CPU Bottleneck, Could you post more about how you configured the devices and what SFP module you used?

Hi,

Bandwidth scenarios

1 - Internet throughput
PC with 2.5GB NIC => port SFP+ #2 (S+RJ10) Mikrotik CSS610 SFP+ #1 (FS SFP+ module @10Gb mono mode 1310nm) => port SFP+ (FS SFP+ module @10Gb mono mode 1310nm) RB5009 ether1 (2.5 GB baseT) => ITSP provider modem (2.5 Gb Base T)

max throughput : 1.5 Gbps with cpu speed to auto
max throughput : 1.9 Gbps with cpu speed to 1400 Mhz

Same config with UDM SE instead of RB5009 : 2.5 Gbps

2 - Local throughput (internet is shutdown)
removing ether1 from WAN list interfaces
adding ether1 to localbridge (LAN)
PC with 2.5GB NIC => port SFP+ #2 (S+RJ10) Mikrotik CSS610 SFP+ #1 (FS SFP+ module @10Gb mono mode 1310nm) => port SFP+ (FS SFP+ module @10Gb mono mode 1310nm) RB5009 ether1 (2.5 GB baseT) => Openmediavault NAS server (2.5 Gb Base T)

results (SMB/iperf3) : 2.5 Gbps throughput

How do you receive a WAN IP (PPPoE or DHCP, VLAN)?
Also is allow fastpath in IP-Settings enabled, also try deleting all Firewall rules and test maybe it is a CPU Bottleneck...

Check tool/profile and set to all
If one of the cores is at maximum, there is a bottleneck on CPU.

Hi,

Ok I know what's going on now.

WAN 1.5 Gbps or more can be reached only if WAN port (or vlan) is on SFP+ port !

If your WAN (1.5 Gbps +) is on ether1, you can only have 1.5 Gbps.

Bug ? How can I report it ?

I can reach 2.5Gb on ether1 (I use it as VLAN trunk to AX3, ether1).
In your faulty case, is speed on ether1 being advertised as 2.5Gbps ? If not, it makes sense it doesn't use full speed ...
So (assuming 2.5Gbps is being advertised for you) a possible difference I see is that in my case ether1 is part of the bridge, in your case it is not (I assume).
If you look at block diagram of RB5009, all ports incl. SFP port go to the same Marvel switch chip so they all should behave the same (given speed constraints per port).
Odd ...

If you really think it is a bug, set faulty config again, make supout.rif and send to support.
Include sufficient explanation what you tested and what you see.

Thanks,

I was able to get 2.5 Gbps on ether1 only when I add this port on localbridge, (i.e as a LAN port) => No problem
for wan purposes (excluding ether1 from localbridge), max bandwidth 1.5 Gbps - (weird thing 1.9 Gbps if I settle CPU to 1400 Mhz)

And you are certain this is not due to some config issue ?
Check tool profile / all while testing the faulty config. If one of the CU cores gets close to 100%, you have a bottle neck.
Because if you remove that port from the bridge, I understood it becomes routing and that's handled by CPU.

But it still doesn't explain then why it can be done using SFP since it follows the same principles.
Supout and support

Certain indeed

load was low on all CPUs

Hi,

i have a similar problem.
The bandwidth on my Wireguard VPN (RB5009 as client) is much slower with CPU Frequency on auto.
CPU on auto: ~130Mbit/s
CPU on 1400MHz: ~270Mbit/s

On auto i've observed that the frequency stays of 700MHz most of the time.

I have an open support case with Mikrotik about this. It has been open for months.

They have confirmed that this is a problem on all RB5009 devices and they do not have an ETA for a fix. The last time I checked, it was not even a bug being worked in the testing and development channels.

We're screwed.

Support told me to try a CRS310 instead.