RB450G slow throughput

RouterOS General
1

I have a RB450G which is setup in router mode utilizing the first port as the Internet facing side and the other 4 ports as the local side.

When I pass traffic from the Internet to any of the four ports locally I’m getting the full speed I’d expect (20meg). However, when I try to pass traffic from one of the four local ports to one of the other local ports by traffic dumps down to about 1.5meg/second.

I’ve checked for errors on the ports and there are none. All ports are set to 100-Full on both the RB450G and the equipment’s side.

Is this a backplane speed issue of some sort I’m running into, or something else?

AGAIN:

Internet (port1) <—> Ports2-4 - I get a full 20Meg/second
Ports2-4 <---->Ports2-4 I get about 1.5meg/second

2

Are you using the default config or did you roll your own?

If you rolled your own and didn’t properly clear the default config, that could be your problem.

BTW A proper clear is “system reset-configuration” from the CLI. You can’t properly clear it from Winbox unless you open a terminal and run that command.

In the default config, I believe they slave ports 3,4,5 to port 2 using the switch chip.

Tom

3

I did clear. However, i have it slaved as you described above. Is that an incorrect way to set it up?

1 is Internet
2-5 are local.

4

My Interface configuration is as follows:

0 R name=“Level3” mtu=1500 l2mtu=1524 mac-address=00:0C:42:AC:58:84 arp=enabled
auto-negotiation=no full-duplex=yes speed=100Mbps master-port=none
bandwidth=unlimited/unlimited switch=switch1

1 R name=“Wireless-Backhaul” mtu=1500 l2mtu=1524 mac-address=00:0C:42:AC:58:85
arp=enabled auto-negotiation=no full-duplex=yes speed=100Mbps master-port=none
bandwidth=unlimited/unlimited switch=switch1

2 R name=“Wireless-BCTech” mtu=1500 l2mtu=1524 mac-address=00:0C:42:AC:58:86 arp=enabled
auto-negotiation=no full-duplex=yes speed=100Mbps master-port=none
bandwidth=unlimited/unlimited switch=switch1

3 R name=“Mailserver” mtu=1500 l2mtu=1524 mac-address=00:0C:42:AC:58:87 arp=enabled
auto-negotiation=no full-duplex=yes speed=100Mbps master-port=none
bandwidth=unlimited/unlimited switch=switch1

4 R name=“Lock-Haven” mtu=1500 l2mtu=1524 mac-address=00:0C:42:AC:58:88 arp=enabled
auto-negotiation=no full-duplex=yes speed=100Mbps master-port=none
bandwidth=unlimited/unlimited switch=switch1

NAME BYTES PACKETS DROPS ERRORS

0 R Level3 10306734431/1489704771 10694299/8325876 8/0 0/0
1 R Wireless-Backhaul 6803412092/13706684825 11489317/13916119 475/0 1/0
2 R Wireless-BCTech 221213209/210989579 1422631/1542456 0/0 361/0
3 R Mailserver 206139945/129922531 449444/959537 0/0 0/0
4 R Lock-Haven 151699657/2272053147 1156191/2282139 0/0 6130/0
5 R bridge 7167862352/16214594731 14481848/17029805 0/0 0/0

The BCTech and Lock-Haven errors I’m pretty sure are a port negotiation issue (I’m working on those). The issue I’m having is running from ‘Wireless-Backhaul’ to ‘Mailserver’. Both of which are set to Full-100 on both sides.

I just noticed that the Level3 interface is set to switch=switch1. Is that correct? Or should that be taken out since it is the external router port.

I also just realized I’m bridged:
[admin@MNFDPA-MTRTR-HUT] /interface bridge port> print
Flags: X - disabled, I - inactive, D - dynamic

INTERFACE BRIDGE PRIORITY PATH-COST HORIZON

0 Wireless-Backhaul bridge 0x80 10 none
1 Wireless-BCTech bridge 0x80 10 none
2 Mailserver bridge 0x80 10 none
3 Lock-Haven bridge 0x80 10 none

If I understand this correctly, I should kill the bridge, and make sure 0-3 are switched, correct?

5

Yes kill the bridge and please report the results.

When you slave ports, you are using the switch chip hardware to create a layer 2 switch. The CPU is not used. When you bridge, you are using the CPU to do the same thing. I suspect you are creating a loop where the CPU and the switch chip are fighting each other.

Tom

6

Just did this this morning… seeing much better throughput!

The only question remaining is about the WAN Router port… does that stay in the ‘switch1’ group? I noticed if I take it out it defaults to half-duplex, and I can’t set it to full-duplex.

7

The WAN port should not be in a switch group. You should let it auto detect speed and duplex.

Tom

8

Thanks for all of the input so far. Unfortunately, it appears in switch mode I lose the complete ability to see distinct statistics for each port and can’t run CALEA. That won’t fly.

So, going back to bridge mode, do I need to yank the ports off the switch? Or do they stay? Eg when port 1 is a member of bridge1 should it also show up under switch1? If not how do I remove it? I didn’t see anywhere to remove a port from the switch.

9

You have to remove the hardware switching to bridge in software. To do that, go to “/interface ethernet” and set the master-port of all interfaces to “none”.

Throughput will be severely affected.

10

Why do you say it will be severly limited? I have a few RBs running bridged and we’ve seen no problems.

So then I did everything right, but the ports still show associated to “switch1” even though there is no longer a master port linking. Is that normal?

11

Probably. “switch” will always show the switch chipset they belong to, the master port property shows if the switch chip is actually used. Post the output of “/interface ethernet print detail” if you need confirmation.

Bridging in software is a lot slower than switching in hardware. You cannot bridge at gig speeds in software, the hardware chipset will easily switch at gig speeds.

Whether the software bridging performance is sufficient for you depends on your environment and what you’re trying to do, and what work the router has to perform. If it’s also busy doing firewall filtering or QoS it will have fewer CPU resources available to bridge packets in software.

12

Here’s what I’ve got, which I believe is correct, but just wanted confirmation:

Flags: X - disabled, R - running, S - slave
0 R name=“Level3” mtu=1500 l2mtu=1524 mac-address=00:0C:42:AC:58:84 arp=enabled
auto-negotiation=no full-duplex=yes speed=100Mbps master-port=none
bandwidth=unlimited/unlimited switch=switch1

1 R name=“Wireless-Backhaul” mtu=1500 l2mtu=1524 mac-address=00:0C:42:AC:58:85
arp=enabled auto-negotiation=no full-duplex=yes speed=100Mbps master-port=none
bandwidth=unlimited/unlimited switch=switch1

2 R name=“Wireless-BCTech” mtu=1500 l2mtu=1524 mac-address=00:0C:42:AC:58:86 arp=enabled
auto-negotiation=yes full-duplex=yes speed=100Mbps master-port=none
bandwidth=unlimited/unlimited switch=switch1

3 R name=“Mailserver” mtu=1500 l2mtu=1524 mac-address=00:0C:42:AC:58:87 arp=enabled
auto-negotiation=no full-duplex=yes speed=100Mbps master-port=none
bandwidth=unlimited/unlimited switch=switch1

4 R name=“Lock-Haven” mtu=1500 l2mtu=1524 mac-address=00:0C:42:AC:58:88 arp=enabled
auto-negotiation=no full-duplex=yes speed=100Mbps master-port=none
bandwidth=unlimited/unlimited switch=switch1

I don’t need Gig speeds, only 10/100. I find it to be a HUGE oversight on the part of MikroTik/RouterOS that when I enabled switching only (no bridge) I lose visibility of graphing throughput on each port. Further, I can’t run CALEA functions. Both of those problems prohibit me from running in true switch mode. But at the moment I’m barely using 2% CPU, so I don’t think there’s any reason to be overly concerned.

13

“I find it to be a HUGE oversight on the part of MikroTik/RouterOS that when I enabled switching only (no bridge) I lose visibility of graphing throughput on each port”

Ofcourse, its because the switch is HW switch so once you enable it, it doesnt go through the CPU and doesnt slow it down. If you dont ilke that, theres always SW bridge which is “SW switching” performed by CPU.

14

So either the “software switch” which is supposedly “way too slow to use” is the same thing you get with a Cisco switch or an HP switch, which allows you to graph… or the HW switch implementation in MikroTik is faulty.

15

Theres something terribly wrong with your setup, this board has quite powerfull CPU which is able to bridge at least 600MBps without conntrack or QoS using big packets, no way its only 1.5MBps…

16

Thanks… you’re the first breath of sense I’ve seen in this conversation. As far as I knew, the bridge was fine to run traffic through, but I wanted to get confirmation on that.

As far as our issue. It seems to be something odd with routing to one subnet we have. I haven’t figured out what’s going on, but the bridge doesn’t seem to be playing into it.

We’re back running bridged and everything is fine.

17

Hi,

I have updated my RB450G’s RouterOS from 4.10 to 5.1 and several hours after I realized that LAN throughput is very slow, the same speed mentioned in the first post of this topic. Before the update the speed was greater than 400 Mbits/sec with the same configuration.
I read all posts and tried resetting the router’s configuration to it’s default, but nothing. Then I disabled the bridge and set up hardware switch. After that the speed was about 220 Mbits/sec between my two computers that use 1 Gbps connection, but one of the ethernet ports became unusable, but I need all port.
I have tested the network hardware of my machines by connecting them directly to each other and the speed was about 700 Mbits/sec so the network cards and cables are working properly.
The speed test was taken with iperf and bwtest with very interesting results:
iperf:

------------------------------------------------------------
Client connecting to 192.168.1.50, TCP port 5001
TCP window size: 8.00 KByte (default)
------------------------------------------------------------
[128] local 192.168.1.2 port 53443 connected with 192.168.1.50 port 5001
[ ID] Interval       Transfer     Bandwidth
[128]  0.0-10.0 sec  5.25 MBytes  4.40 Mbits/sec
[128] 10.0-20.0 sec  5.78 MBytes  4.85 Mbits/sec
[128] 20.0-30.0 sec  10.9 MBytes  9.14 Mbits/sec
[128]  0.0-30.1 sec  21.9 MBytes  6.11 Mbits/sec

During iperf test the rotuer’s CPU load was about 8% so the bottleneck is not the CPU.

bwtest between the router and my PC (router was the sender): ~ 330 Mbits/sec (UDP); ~31 Mbits/sec (TCP)
bwtest between the router and my PC (router was the receiver): ~58 Mbits/sec (UDP); ~17 Mbits/sec (TCP)

What should I do to speed up the router?

** My configuration (as fewi requested in his signature) **

/ip address print detail 

 0   address=192.168.1.1/25 network=192.168.1.0 interface=ether2 actual-interface=bridge 
 1 X ;;; VPN
     address=192.168.10.1/24 network=192.168.10.0 interface=ether1-gateway actual-interface=ether1-gateway 
 2   address=192.168.10.2/32 network=192.168.10.0 interface=ether2 actual-interface=bridge 
 3 D address=178.xx.xx.xx/22 network=178.xx.xx.xx interface=ether1-gateway actual-interface=ether1-gateway




/ip route print detail 

 0 ADS  dst-address=0.0.0.0/0 gateway=178.xx.xx.xx gateway-status=178.xx.xx.xx reachable ether1-gateway distance=0 scope=30 target-scope=10 
        vrf-interface=ether1-gateway 
 1 ADC  dst-address=178.xx.xx.xx/22 pref-src=178.xx.xx.xx gateway=ether1-gateway gateway-status=ether1-gateway reachable distance=0 scope=10 
 2 ADC  dst-address=192.168.1.0/25 pref-src=192.168.1.1 gateway=bridge gateway-status=bridge reachable distance=0 scope=10 
 3 ADC  dst-address=192.168.10.0/32 pref-src=192.168.10.2 gateway=bridge gateway-status=bridge reachable distance=0 scope=10




/interface print detail 

 0  R  name="ether1-gateway" type="ether" mtu=1500 l2mtu=1526 
 1  R  name="ether2" type="ether" mtu=1500 l2mtu=1524 
 2  R  name="ether3-pc" type="ether" mtu=1500 l2mtu=1524 
 3     name="ether4-wrt" type="ether" mtu=1500 l2mtu=1524 
 4     name="ether5-nas" type="ether" mtu=1500 l2mtu=1524 
 5     name="pptp-anyuka" type="pptp-in" 
 6     name="pptp-tomikaa" type="pptp-in" 
 7  R  name="bridge" type="bridge" mtu=1500 l2mtu=1524
18

My RB450G with RouterOS 5.4 also has throughput problem, the WAN to LAN NAT throughput only get 200Mbit/s while using iperf, the CPU utilization reach 100%, what should I do?

19

200 Mbps is the maximum for 450G with NAT

20

I have a 100 mbit wan connection. When I use pppoe over Windows dial than I got the full 100 mbit. With the routerboard RB450G I got only 48 mbit WAN routing throughput with the default factory configuration with os 5.15. Whats wrong? The CPU load peak is after a 10 minutes test only by 35 % and the average load is by 25%.