Hi,
i have this OSPF scenario (see attached image) and a main problem to solve.
I want to force traffic flow using interface cost and the only way to do that, is enstabilish an EOIP tunnel between R1-R2 with a /30 p2p and put this network on OSPF Area-1, otherwise traffic generated from client on R2 goes through R3 → R4 → R1 and not directly to R1. The reason why this happen is that i use redistribuite-connected on intances because my clients have PPPoE and OSPF announce /32 address on Area-1, so traffic chose the “worst” path because OSPF always prefer inter-area path instead of intra-area path and don’t consider costs of interfaces.
Am i wrong?
Now, with EOIP configured, i haven’t particular problems, except of some (random?) cases, some client has interruptions during download/upload (HTTP,FTP,IPTV, ecc) and i think that problem is due to MTU or fragmentation on EOIP (EOIP MTU is manually set to 1500).
Is there other possibilities to have this OSPF scenario and don’t have any issue without using EOIP? Changing EOIP MTU or DONT FRAGMENT parameter can solve my issue?
I suppose you announce customer /32 in area1 and not in the backbone because you want to limit the backbone database size.
If this is not true, how are the /32 announced in area1 and how are they announced in the backbone? Can you post the corresponding /routing ospf lsa print ? Area 0 should be preferred to area nonzero.
If this is true, then traffic has to stay in area 1. I would consider:
best option: creating a VLAN on the backbone interface between R1 and R2, if it is ethernet, numbering it and putting it in OSPF area 1, or
ugly workaround: increasing the physical MTU in the whole backbone, so you can use eoip or GRE without fragmentation
First things first, don’t use any tunnels to force routing protocols work as you want. Just read some docs, relax and do some work.
And now. Does R1 ospf router redistribute default route? Is it ASBR? If yes, then on R2 you should see external E2 routes from R1 in lsa type 5. If all your routers in same area, they should calculate shortest path between each other and R1 must be shortest route for R2 to E2 routes. What can you see in lsdb and route list?
And good advice. Do not use OSPF as customer’s service routing protocol. Use BGP or MPLS+MP-BGP.
I can’t find a way to announce /32 on a determinate area, OSPF do all itself, because /32 assigned to the clients are dynamic and it’s not given from local POOL but by RADIUS Server directly. How can i select an Area to announce /32 dynamic networks, without adding manually on /routing ospf network, is it possible?
R1 and R2 are connected by wireless p2p and you’re right, i can try to setup a VLAN (i can’t believe that i don’t think this yet), i will try
Don’t redistribute /32 routes in OSPF. It is not scalable. With each connected client ospf routers will have to recalculate their database - bad idea. Even with BGP it’s bad idea with endless updates. You shouldn’t redistribute /32. Give out /24 or /16. And if client connecting to it’s pppoe server, just build for him pppoe session without routing.
I must work on a network of an WISP that grown up so fast and without a real standard for configuration and network topology. I spent some nights to try configuration to obtain what i need.
I want “force” traffic flow because link between R3-R2 (in this example) must remain a backup.
This is a small part of our network, we have at least 130 router and piece by piece i would like to simplify OSFP topology to run better.
I haven’t a PPPoE concentrator configuration and i give to my client a PUBLIC IP ADDRESS by PPPoE but i can’t have a POOL of PUBLIC IP for each router, so i can’t summarize networks.
I know that is not the best solution.
Anyway i tried to setup a VLAN with /30 running on OSPF Area1, trafflic flow is right and seems don’t have any issue with fragmentation, i tried to ping with NO FRAGMENTATION above 1500bytes.
Now i’m waiting a response from client that experienced download/Upload interruption.
I red some docs, i do MTCRE, i read every day forums and try to understand as best as i can what i’m trying to do before doing it.
R1 don’t have redistribuite dafult, it isn’t our master router before going to the internet.
I dont know if you see image, but trough R2 and R1 pass the backbone ring and at sametime R1 and R2 are connected to the same area that include R3-R4
Now I get it Sorry for misunderstanding your topology.
If you want to force traffic go to R1 from R2, just set ospf cost higher, don’t mess with tunnels. Set on link from R2 to R3 ospf cost 10 and you’ll be fine.
With thia particular/wrong configuration, OSPF ignore interface cost, beacuse (i think it s inner of OSPF algortim) the /32 connected rouetes is announced on Area-1 and not to backbone, so the traffic goes to R3->R4 beacuse is considered inter-area traffic. If i put whatever network on routning-ospf-networks i can force Area and dosen’t have any problems, but for reason just sayd, i can’t do this and i must find a trick to bypass the problem.
You wrong about this: “so traffic chose the “worst” path because OSPF always prefer inter-area path instead of intra-area path and don’t consider costs of interfaces.”
You should be able do use /routing ospf network add area=1 network=<the whole big prefix containing the /32s>. But I still not understand: if you announce them the same way in both areas, area 0 should be preferred. Why and how exactly are you using a multi area setup?
If you inherited a network and you cannot or do not want to change it too much, I suggest you add a VLAN in the backbone link and put it in area 1.
Sorry for misunderstanding your topology.