The attached diagram shows a segment of my network that I'm having trouble with.

Yellow IP addresses are Loopback IP's attached to a Bridge called 'loopback'. (as per 'best practices')

The two RB450g's in the middle of the diagram are running bridges - called Backhaul - with the interfaces that connect to the M5's in that bridge (to minimize hop count).

ALL the RB450g's in the diagram are running OSPF. The two 450g's that are configured with Backhaul bridges are running OSPF on the IP that is attached to the Backhaul Bridge interface. They need to do OSPF because they attach to other segments.

All three RB450g's that are running as Routers have alternaet path to 0.0.0.0. However 10.255.255.33 is the preferred path due to latency and Bandwidth available at that end followed by 10.255.255.20, and lastly 10.255.255.45.

All RB450g's are running 5.16. All are running NTP synced to the same time source. All running Simple Authentication (not MD5).

All the UBNT M5's are in Client/AP-Bridge mode. (NOT WDS)

10.255.255.33 is DR (priority 99), and 10.255.255.20 is BDR (priority 85) and 10.255.255.45 is priority 75. The other two RB450g's are priority 0. All RB450g's are in Area 0. The are no ASBR's in this diagram.

I've tried both NBMA and Broadcast mode for OSPF.

In NBMA mode I never achieved convergence on ALL routers at the same time - the neighbors tab would show at least 2 of the 5 RB450g's with a 'router ID' of 0.0.0.0. State changes would not propagate, or if they did they were VERY slow.

Myself and two people checked the configurations of all the RB450g's at least 10 times. There was no reason why convergence should not have happened.

In Broadcast I'm seeing dozens of state changes an our as the multicast's get 'lost' (which is not unexpected I guess), causing breif periods of 'icmp unreachable' responses to running pings.

THIS TOPOLOGY WORKED before the link to the router at 10.255.255.45 was added a couple weeks ago - that required a Ethernet interface be added to the Backhaul Bridge at 10.255.255.65. NMBA started to fail then. I switched to Broadcast mode a couple nights ago because in NBMA mode the default route was disappearing at random times for 10 to 120 minutes at a time - often requiring one or more RB450g's to be rebooted.

I use NBMA everywhere else in the network - with similar topology, but this is the only segment where there are two RB450g's 'back to back' running as Bridges...

I'd REALLY rather not eliminate the bridges but I'm about ready to do just that... although I don't know if it'll help.

I'm open to suggestions

I use ospf in my network (100% MT) and do not use loopback or NBMA

http://wiki.mikrotik.com/wiki/Manual:OSPF-examples

There are three basic elements of OSPF configuration:

Enable OSPF instance
OSPF area configuration
OSPF network configuration

PTP's Example
/routing ospf area
set backbone area-id=0.0.0.0 disabled=no instance=default name=backbone type=\
default
/routing ospf network
area=backbone network=10.100.0.20/30 ( for wlan PTP )
area=backbone network=10.100.0.24/30 ( For Ether PTP)

PTMP's Example
/routing ospf area
set backbone area-id=0.0.0.0 disabled=no instance=default name=backbone type=\
default
add area-id=10.100.99.0 default-cost=1 disabled=no inject-summary-lsas=yes \
instance=default name=pppoe type=stub

/routing ospf area range
add advertise=yes area=pppoe cost=calculated disabled=no range=10.100.99.0/24

/routing ospf network
add area=backbone disabled=no network=10.100.0.24/30
add area=pppoe disabled=no network=10.100.99.0/24


Which works OK for me

So you are not using M5's for transport then but using MT's for the PtP links ? Are you bridging or routing the PtP's ? I had ALL MT a couple years ago but had to moved to M5's because, at the time, NV2 was very poor, and I couldn't get anymore bandwidth out of Nstreme / Nstreme Dual.

This topology can handle ~50Mbps with no change in latency, and 80Mbps is the highest I've seen - but latency was 4 x at that point.

I disabled connection tracking on all the RB450g's long ago to increase throughput.

All the M5's are running v5.3.5.

I can post configs if it's helpful. Myself and two other admins have all gone over them multiple times so I'm pretty confident they are correct, and we do this elsewhere in the network too so it's not new to us, but with 5 OSPF configs it will be a pretty long post...

Thx
Z

Not using M5's, network is routed, NV2 latency is a concern but has it's advantages

Our network is large enough that a lot of links have to be collapsed to minimize hop counts. All the routers in the diagram connect to other bridged segments which are configured the same and work fine... anyways I gave up and installed two switches in place of the MT's in Bridge mode. Changed nothing else, and the AS has been stable for 36 hours, which is 35.75 hours longer than previous.

Special....

All the M5's are running v5.3.5.

I have a pair of M5's back hauling a segment of my network and was receiving constant ospf error. Nothing I tried seemed to help. When UBNT release v5.5 I read the release notes and one of the problems it fixed was a hanging problem. I upgraded the problem M5 link and the ospf errors disappeared. I have had 0 errors for almost 20 days.

Thanks for the heads up. I haven't tried AirOS 5.5 anywhere yet... maybe that's a fix too.

I have other segments like this where the rockets plug into switches instead of RB's and the OS's are the same on each platform as on this segment.... The difference here was that I wanted to skip installing network switches and just use the RB450g with bridging on two Ethernet ports.

It's worked fine since I put in switches, connecting the M5's to them and the RB450g with the former Bridge Interface IP of the 450g now assigned to an Ethernet interface on the RB instead.