Best method of load balancing / bonding with Netonix?

[agaftcomms]

I have 3 possible methods for load balancing and i'm unsure of the best method to use with a Netonix switch

The network topology is

Code: Select all

 
.........................AP-A ~~~~~~~~~~~~~~~ Station-A
......................../...........................\
Router<--->Netonix Switch .......................... Netonix Switch<--->Router
........................\ ......................... /
.........................AP-B ~~~~~~~~~~~~~~~ Station-B

1) Use LACP on the Netonix switches and let them load balance everything
2) Use VLAN101 for Link A & VLAN 102 for Link B. Use MSTP on the Netonix Switches. Put both VLAN's into a bonding group on the routers
3) Convert from bridging to routing on the Links
4) Something else

I would use LACP if this was ethernet to ethernet, but I also need management access to the wireless radio's. My preference would be to use LACP but my question is how do I get to the radio's? LACP will split the traffic according to MAC/IP/Port but it is unpredictable as how to which port it is going to use, so I can't accurate predict that traffic to AP-A and Station-A will go out via the appropriate ethernet port
I thought of using different management VLAN's and only having i.e. VLAN101 on port 5 and VLAN102 on port 6 so it ensures traffic to each set of radio's only goes through that exact link, but when I configure different VLAN's I lose connectivity. I presume this is because LACP is seeing a different config per port and dropping them out of the LAG group?

The 2nd option seems pretty good but MSTP is in beta right? This is for a production network so it has to be reliable

Option 3 is possible but not with the radio's I have in place

[mike99]

MSTP priority don't work so you can forget it for now.

OSPF load balancing won't be more predicatable than LACP. With static or policy routing, you could route a subnet via link a and a other an other but predictable mean less effective since routes won't change automaticly if a link is overload or a radio is down (router won't be able to see if link is up or down since it pass through a switch so metric won't work).

I would go with LACP.

MSTP would be a great option if you would like to pass data through link A and VoIP through link be to be sure your VoIP link is never saturate. Anyway, we will have to wait for Eric to fix it or fix ERPS that could do the same thing with faster convergence (but need at less 3 swtichs in ring and you have only 2 switchs in your schema so ERPS could not be use in your case).

[agaftcomms]

A routed network can be load balanced with PCC, ECMP or BGP so that's not a problem, it's just a lot more messing around to set up and i'd probably have to change the radio's

I agree LACP would be fine however.... won't it cause an issue with actually reaching the radio's for management? LACP works great on ethernet, and you don't really care which link path the data takes. But when you are using radio's in between you need the data to take a specific path to reach them. Hence why I thought to use separate VLAN's for management but if I exclude the other link from that VLAN I seem to lose connectivity entirely (presumable due to LACP not liking different config on each port)
Or is netonix smart enough to work out that it can only possibly reach a particular MAC address through a certain link and hence I can just put the one management VLAN on both sides?

[mike99]

Argh, just lost my answer because I was not logged anymore so I will answer fast this time.

Never say it can be done via layer 3, only that OSPF ECMP won't be more predictable than LACP. PCC seem to be a Mikrotik name for policy routing with ping to know if gateway is up and I think it could be a good L3 option by using different metric. iBGP is overkill but it support TE or maybe filtering to block the management VLAN from the other link, probably via route-map option (not sure for Tik). MPLS + TE could be also an option but most radios won't be able to read the EXP bit to apply QoS.

For you concern with LACP, I'm not sure how it will react but what I would try is 2 differents management VLAN for each link, management vlan exclude on the other link, to see how LACP will react (if it will only load balance vlan tagged on both link).

[agaftcomms]

For you concern with LACP, I'm not sure how it will react but what I would try is 2 differents management VLAN for each link, management vlan exclude on the other link, to see how LACP will react (if it will only load balance vlan tagged on both link).

Already tried this, as mentioned I can't set different exclusions for VLAN's on the interfaces otherwise I completely lose connectivity on that link - and i'm guessing its because of LACP seeing a different config on each interface and thus dropping the interface out of the group

Given that Netonix is designed specifically for the WISP community and this kind of topology is extremely common there MUST be a way to do this properly..... Can I get a Netonix rep to chime in?

[mike99]

I read a little more about this and seem like LACP is not VLAN aware and must be use on trunk port that will pass all VLAN. The protocol was made with switch to switch communication in mind, not for use with bridged device between them. It can be done over bridge but management access to those devices will be a problem.

The best option that come to mind without layer 3 option would be to use radio with dual ethernet port, one for traffic and one for data. I also looked at 2+0 link aggreagation doc from SAF and 2 ethernet port seem the way to go. If I look at SAF doc for 2+0, seem like both need to have a ethernet port dedicated to management (look at the image at page 8):
https://www.saftehnika.com/files/downlo ... %20Oct.pdf

I don't know what radio you use but I think all models of airfiber have a data port and a optionnal management port. That would be a good radio for LACP. Else, OSPF ECMP with static route for management access seem to me the best layer 3 option.