Thanks Spudrageous,
Two reactions - (not criticisms!):
1. Most OSPF tutorials and most of the (lesser # of) deployments I've seen ignore redundancy! I.e., they have a single ABR between each area and Area 0. We are very focused on no single point of failure. For this, you can stick two ABRs between area 0 and area X but then, to steer bidirectional traffic over a preferred path (say fiber vs radio), you need to configure an area X tunnel through Area 0 between these two ABRs. Any discussion of OSPF areas should include a discussion of redundancy but it's seldom mentioned. That said, for historical reasons we ended up with four separate area zeros as private ASNs interconnected by eBGP and upstream with BGP confederation. So, we ended up with complexity in another form...
2. I can attest to modern routers easily handling larger networks. We have areas with 20 and 60 routers, but we also have one area with 147 routers and over 3000 OSPF routes. The routers are CCR1036, CCR1009, RB4011, plus a few residual RB3011 & RB2011 devices. In the 25+ years people have been deploying OSPF, router processors have gone from an 80 MHz PowerPC to quad ARM cores at 1.9 GHz with hardware off load (on the RB4011)! That's a non-trivial upgrade.
Our network with 147 routers runs one OSPF area zero with no problems. Adding a route at one edge is reflected on a router at a distant edge (8 hops) within one Winbox refresh interval (1 sec?) and we can force route changes by marking a core link passive and see no packet loss at a downstream client site.
That said, our stability may be helped by a script we developed in 2012 (for RouterOS 4) which still runs once a day at 6am on every router. It deletes (if present) and then recreates a set of high-cost static routes duplicating the current OSPF routes. This was critical for stability of our original OSPF network on RouterOS 4 on RB750UP routers in 2012. I haven't been able to prove these backup routes are ever invoked today, but we haven't deleted the script.
I'm definitely interested in hearing about other's experiences.
Two reactions - (not criticisms!):
1. Most OSPF tutorials and most of the (lesser # of) deployments I've seen ignore redundancy! I.e., they have a single ABR between each area and Area 0. We are very focused on no single point of failure. For this, you can stick two ABRs between area 0 and area X but then, to steer bidirectional traffic over a preferred path (say fiber vs radio), you need to configure an area X tunnel through Area 0 between these two ABRs. Any discussion of OSPF areas should include a discussion of redundancy but it's seldom mentioned. That said, for historical reasons we ended up with four separate area zeros as private ASNs interconnected by eBGP and upstream with BGP confederation. So, we ended up with complexity in another form...
2. I can attest to modern routers easily handling larger networks. We have areas with 20 and 60 routers, but we also have one area with 147 routers and over 3000 OSPF routes. The routers are CCR1036, CCR1009, RB4011, plus a few residual RB3011 & RB2011 devices. In the 25+ years people have been deploying OSPF, router processors have gone from an 80 MHz PowerPC to quad ARM cores at 1.9 GHz with hardware off load (on the RB4011)! That's a non-trivial upgrade.
Our network with 147 routers runs one OSPF area zero with no problems. Adding a route at one edge is reflected on a router at a distant edge (8 hops) within one Winbox refresh interval (1 sec?) and we can force route changes by marking a core link passive and see no packet loss at a downstream client site.
That said, our stability may be helped by a script we developed in 2012 (for RouterOS 4) which still runs once a day at 6am on every router. It deletes (if present) and then recreates a set of high-cost static routes duplicating the current OSPF routes. This was critical for stability of our original OSPF network on RouterOS 4 on RB750UP routers in 2012. I haven't been able to prove these backup routes are ever invoked today, but we haven't deleted the script.
I'm definitely interested in hearing about other's experiences.
Statistics: Posted by Brough — Thu Jan 04, 2024 10:54 pm