[c-nsp] wisdom of installing many routes into ospf

Mike mike-cisconsplist at tiedyenetworks.com
Sun Aug 29 23:28:09 EDT 2010


Pshem Kowalczyk wrote:
> Hi Mike,
>
>
> On 29 August 2010 12:30, Mike <mike-cisconsplist at tiedyenetworks.com> wrote:
>   
>> Hello,
>>
>>   I have a BRAS terminating many pppoe sessions (>2000 lets say). I would
>> like to allow my bras to participate in OSPF and I am wondering what the
>> collected wisdom about doing so is. I want to be able to assign some of
>> these sessions fixed subnets for example, and this is one motovator. Another
>> is that even for connected interface addresses, I sometimes want special
>> routing (eg: instead of core -> bras -> end user, I may want core ->
>> intermediate -> bras -> end user) which I can use ospf to override. Has
>> anyone done this and has anyone any references where I might see an example
>> or some documentation on pitfalls and so forth?
>>     
>
> A few things can affect you're outcome (especially the stability of
> the PPPoE sessions) but generally I would advise against using OSPF in
> this scenario. The main reason is the fact that every single time a
> session is established or dropped the whole OSPF area has to re-run
> SPF algorithm, depending on the hardware involved this might or might
> not be a big problem. Next thing is scalability - if you expect
> significant growth in number of routes - I'd suggest you move to BGP,
> OSPF is likely to get slower as number of routes increases. The last
> reason is ability to control routing - we use BGP communities to mark
> various prefixes so the other devices can make a decision how and
> where to re-advertise particular routes, OSPF gives you some abilities
> as well, but not as extensive as BGP.
>
> We run a setup with multiple BRASes, some with over 40k subscribers.
> We use ip pools on the devices, with dynamic IPs allocated from those
> pools. BRASes only advertise the aggregate pool routes to limit the
> number of prefixes seen by the rest of the network. Only customer that
> have static IPs or networks get 'advertised'  into the rest of the
> network. We use BGP between the BRASes and the rest of the network,
> our BRAS is a PE and for 'special' routing needs we place the PPP
> sessions directly in to the VRF that provides the expected
> connectivity.
>
>   

Thank you for describing your setup. I was originally thinking I'd 
advertise the clients into a special area not the backbone, and 
advertise only aggregated routes as you describe, but after thinking 
some more about it you could be right that BGP may be a better way to go 
due to better filtering and so forth.

Mike-


More information about the cisco-nsp mailing list