Community discussions

MikroTik App
 
sloth
just joined
Topic Author
Posts: 20
Joined: Fri Nov 11, 2011 6:13 pm

RB450G throughput / performance issues

Wed Jan 25, 2012 2:34 pm

The reason I bought the RB450G in comparison to other Routerboard products, is the advertised high throughput.

However, in a test setup CPU usage is at 70-80% with only about 70mbit of WAN to LAN traffic from 1 client with 2 active ftp sessions.
This would mean WAN to LAN throughput would already max out at 100mbit, which is below both of the advertised throughput and my expectations.

Testing procedure:

RB450G with dhcp client on eth1, NAT, dhcp server and hotspot configured at eth5.
On eth5 there is 1 client directly connected to the device, with 1 active dhcp lease and logged in via hotspot. Download speed is tested with 2 active ftp sessions, maxing out the internet subscription at 70mbit.
RB450G has been upgraded to latest RouterOS, version 5.12

Is there anything I can do to improve these performance figures, or is this really only the maximum achievable throughput one can expect from the RB450G?

Thanks in advance
 
mtmx80
Frequent Visitor
Frequent Visitor
Posts: 87
Joined: Wed Mar 16, 2011 11:12 pm

Re: RB450G throughput / performance issues

Wed Jan 25, 2012 3:45 pm

I would try it first with "factory default" settings: eth1 WAN, eth2-5 LAN (eth2 master, eth3-5 slave), DHCP, NAT; nothing else.
This would make you more confident if router is overloaded by any of your additional settings or you are simply reaching it's limits easily.

Note 1.: I can do more with same CPU
Note 2.: A smallest configuration issue may top out that CPU on 100%.
 
sloth
just joined
Topic Author
Posts: 20
Joined: Fri Nov 11, 2011 6:13 pm

Re: RB450G throughput / performance issues

Wed Jan 25, 2012 5:57 pm

Thank you for your reply.

As you suggested, I tried reverting back to "factory default-like" settings:

eth1 WAN, eth2-5 LAN with 5 being configured as slave with DHCP and NAT.

The good news is, using this config CPU usage dropped in half, peaking at ~30% with 60mbit of WAN to LAN traffic using the same speed test as in the first post.
I found this post stating a RB450G with NAT would max out at 200Mbit, which seems on par with the aforementioned results.

However, with this config I am unable to set up hotspot: the server item at IP->Hotspot in Winbox is now red colored and unable to proceed.

It looks like this has something to do with Switch Chip Features but being a novice Routerboard user, I don't fully understand what the relationship between both setting a master port with slaves, hotspot and throughput performance is.
 
brandonrossl
Frequent Visitor
Frequent Visitor
Posts: 61
Joined: Wed Jun 08, 2011 10:09 pm

Re: RB450G throughput / performance issues

Wed Jan 25, 2012 6:45 pm

Master port means what port "controls" the slaves via the switch chip rules.

So if you direct traffic to the master, the switch chip directs it based on the rules that are defined (if any). You can only change behavior on slave ports with switch chip rules (OR don't set a master and use the cpu).

Example I have 433ah

ether 1 is wan with NAT bridged to ether2, ether3 has ether2 set as master and all switching is handled by the switch chip. Ether3 will behave as if it were another port on a smart switch, not a router.
 
sloth
just joined
Topic Author
Posts: 20
Joined: Fri Nov 11, 2011 6:13 pm

Re: RB450G throughput / performance issues

Thu Jan 26, 2012 12:14 pm

So, if I understand the whole switching with master and slaves ports correctly:

Traffic needs to be directed from eth1 to eth5 and vice versa. Switching only works on layer 2, while as eth1 and eth5 are on different subnets, and therefore I cant put them together as both slave ports.
This will also mean there is higher CPU usage, as hardware can't offload.

mtmx80, what kind of throughput do you get on a RB450G, you also use captive portal, right?

Also, right now I am testing with 1 "heavy load" client, downloading at 60mbit whereas in production, clients will be limited to 10mbit download.
Is it less resource intensive handling multiple clients with high aggregated throughput? Are there any benchmarks available to simulate this?

Who is online

Users browsing this forum: No registered users and 20 guests