Community discussions

MikroTik App
 
andryan
newbie
Topic Author
Posts: 36
Joined: Fri Nov 30, 2007 10:33 pm
Location: Jakarta, Indonesia
Contact:

RB333 3.0RC10 ethernet bonding issue

Fri Nov 30, 2007 11:14 pm

Hi all,

I recently bought 2 RB333s (3.0RC10 and no miniPCI) to bond 2 local loops (each via 10Mbps full duplex metro ethernet).
I did some tests before installing them on the network. I tried to simulate 2 10Mbps links and forced both ethernet to 10Mbps full duplex with auto-negotiation off.

Here is how I did my simulation.. I created a balance-rr bonding interface for both ethernet and ran 2 tests (FTP and MT's bandwidth test) with ping continuously running while the tests were running. Performance was very poor when I ran the tests, i.e. lots of RTOs and the transfer speed was very unstable. I tried adding EoIP on top of each ethernet and apply bonding on the EoIPs instead, but that made the performance even worse. I noticed that the CPU usage on both RB333 was not reaching 100%, it wasn't even close so I guessed it has nothing to do with insufficient hardware resources. When I used FTP to test, I was only able to get approx 3000Kbps spread across both ethernet; however when I used MT's bandwidth test, it gave me better throughput. Both tests yielded quite a few RTOs on my continuous ping while they were running.

After my initial tests didn't go as expected, I tried a different scenario. I changed the ethernet settings to 100Mbps full duplex with auto-negotiation on and ran the same tests again. It was capable of delivering 100Mbps full duplex on the bonding interface via both ethernet (each at approx 48Mbps). My continuous ping didn't show any RTO but rather only increased latency (even occasional RTO every once in a while is expected). If it were caused by insufficient hardware power, it wouldn't give such an excellent result when both ethernet are set at 100Mbps full duplex. 10Mbps is one-tenth of 100Mbps and surely that uses less resources, so that doesn't make sense. I believe it has something to do with a bug in the RouterOS.

I hope that the cause of this problem can be located very soon and be fixed immediately since I really need to make use of both my 10Mbps links ASAP. :(

Have anyone else encountered similar problems? I can't compare the results with RouterOS 2.9 since I just own these 2 RB333s.

Thanks!

Regards,
Andryan
 
galaxynet
Long time Member
Long time Member
Posts: 648
Joined: Fri Dec 17, 2004 2:52 pm
Contact:

Re: RB333 3.0RC10 ethernet bonding issue

Sat Dec 01, 2007 4:28 pm

andryan -
I currently use ROS 3.110rc and 2.9.48 for bonded connections - wireless and wired.

While there are a few outstanding issues with wireless on ROS 3.xxrc I have not had any issues with bonding wired interfaces. Use EoIP and add the EoIP tunnels to the bonding interface. Use rr and arp monitor. Make sure your tunnels are setup correctly - MAC addresses and tunnel IDs.......

As to the Ethernet issue - I have forced ROS to use a particular ethernet speed/duplex - it works but poorly. I just leave it at auto and it seems to setup whatever it can and work just fine. So it seems the issue is setting it manually vice automatic selection. So let ROS determine what it can or cannot do automatically and that should take care of that issue.

Thom
Thom Lawless
General Manager
RapidWiFi, LLC
thom.lawless [at] rapidwifi.com
 
andryan
newbie
Topic Author
Posts: 36
Joined: Fri Nov 30, 2007 10:33 pm
Location: Jakarta, Indonesia
Contact:

Re: RB333 3.0RC10 ethernet bonding issue

Sat Dec 01, 2007 5:00 pm

andryan -
I currently use ROS 3.110rc and 2.9.48 for bonded connections - wireless and wired.

While there are a few outstanding issues with wireless on ROS 3.xxrc I have not had any issues with bonding wired interfaces. Use EoIP and add the EoIP tunnels to the bonding interface. Use rr and arp monitor. Make sure your tunnels are setup correctly - MAC addresses and tunnel IDs.......
I believe my tunnels are set up properly. I was able to get both working as expected, it's just that the performance was not satisfactory. :(
As to the Ethernet issue - I have forced ROS to use a particular ethernet speed/duplex - it works but poorly. I just leave it at auto and it seems to setup whatever it can and work just fine. So it seems the issue is setting it manually vice automatic selection. So let ROS determine what it can or cannot do automatically and that should take care of that issue.

Thom
Ah, I haven't tried testing with forced 100Mbps full-duplex with autonegotiation off. If it performs poorly too, then definitely it needs autonegotiation to get decent performance. My metro ethernet terminal cannot do autonegotiation and has to be set manually at 10Mbps full duplex. I have an idea in mind, what if I plug a UTP cable from the metro ethernet terminal to a switch and another cable from the switch to the RB333; then do the same to the other metro ethernet. This way the RB333s will be able to run at 100Mbps full-duplex with autonegotiation on despite only having 10Mbps full duplex link in-between.

Thanks for your reply, Thom. :)
 
galaxynet
Long time Member
Long time Member
Posts: 648
Joined: Fri Dec 17, 2004 2:52 pm
Contact:

Re: RB333 3.0RC10 ethernet bonding issue

Sun Dec 02, 2007 1:26 pm

andryan -
You are welcome.

I read what you said about auto-negotiation and your 'metro ethernet'. You mean to say that if you set the MT for auto-negotiation that it will not 'find' the metro ethernet speed and duplex on it's own? I have a couple of really old AP units (they just refuse to die after 8 years...) that are 10Mbps, half-duplex. The MT backhaul that plugs in to this unit is set for auto-negotiation - it (the MT) correctly identifies the interface as capable of only 10Mbps, half duplex, and so sets itself to 'match' that.

Now the above IS on 2.9.4x ROS, haven't tried that with the 3.xxrc yet (10Mbps / half duplex device) - I was just making sure I understood correctly what you were stating.


Later,

Thom
Thom Lawless
General Manager
RapidWiFi, LLC
thom.lawless [at] rapidwifi.com
 
andryan
newbie
Topic Author
Posts: 36
Joined: Fri Nov 30, 2007 10:33 pm
Location: Jakarta, Indonesia
Contact:

Re: RB333 3.0RC10 ethernet bonding issue

Sun Dec 02, 2007 3:49 pm

andryan -
You are welcome.

I read what you said about auto-negotiation and your 'metro ethernet'. You mean to say that if you set the MT for auto-negotiation that it will not 'find' the metro ethernet speed and duplex on it's own? I have a couple of really old AP units (they just refuse to die after 8 years...) that are 10Mbps, half-duplex. The MT backhaul that plugs in to this unit is set for auto-negotiation - it (the MT) correctly identifies the interface as capable of only 10Mbps, half duplex, and so sets itself to 'match' that.
Yes, well I believe it might work but probably not optimally.

I have previously used a Windows XP PC to connect directly to the metro ethernet terminal at one end and another at the other end. The ethernet driver is set to autonegotiation, it was able to detect 10Mbps but I forgot whether it was detected as full duplex or half duplex. I was only able to achieve ~1Mbps throughput though with this configuration. I forced manual 10Mbps full duplex at both ends and then full 10Mbps throughput was achievable. I'm guessing that it has to be manually configured at 10Mbps full duplex with autonegotiation off. This is probably the limitation of my metro ethernet terminals.

I plan to get a managed switch and create 2 VLANs (since I have 2 metro ethernet local loops) with 2 ports assigned to each VLAN. On this managed switch, one port (which will be set at 10Mbps full duplex) goes to the metro ethernet terminal and the other port to the RB333). This way my RB333s will be 'tricked' into thinking that it is 100Mbps full duplex link with autonegotiation on.
Now the above IS on 2.9.4x ROS, haven't tried that with the 3.xxrc yet (10Mbps / half duplex device) - I was just making sure I understood correctly what you were stating.


Later,

Thom
Ah, hopefully much of the bonding stuff haven't been changed in v3.0RCx, but EoIP throughput performance is still poor in v3.0's RCs.
Looks like you got the idea.. :)

Thanks,
Andryan

Who is online

Users browsing this forum: valentinoo77 and 99 guests