[rbak-nsp] Subcriber route and metric

[Olivier Benghozi]

Hi Blake & Raphael,

>> Welcome to the joys of Redback subscriber routing... I'm not sure what SEOS version you're running or whether their architecture has changed in recent releases, but in the 6.1 trains I'm most familiar with, subscriber routes have the particularity of not being able to load-balance or handle redundancy properly. The behavior is as you describe.  When trying to load-balance across multiple subscriber routes with the same metric, the box will even go so far as to install all three next-hops in the FIB and blackhole traffic to all but the most recent connection.  I observed the same behavior running RIP with the CPE learning three identical routes, but I believe this may have been fixed in recent releases.
>> 
> I'm running 6.1 version also, so I'm exactly in the same situation.
> I don't know if it have been fixed, but a firmware upgrade is just unthinkable (old hardware, old and exotic configuration).

a friend opened a ticket for this "issue", on 6.5, on another network.
Answer: this is by design, WONTFIX, etc.

The last route inserted in the RIB wins.

Except when you activate the loadbalancing between PPP links (8 max), which is available at least in 6.4 : all PPP links are used (hash based distribution).

Of course the "metric" is never considered.


>> We have used a couple of workarounds, none of which are particularly nice but they do get the job done:
>> 
>> - cut your "primary" /24 route into two /25s so that the shortest prefixes will be preferred and you don't have the same set of routes pointing at each subscriber (increases framed-route count by 50% though)-:

It's what we do. It sux a bit, though.


> More general subject, what is your advice about a SE400 lns replacment ?  Cisco ASR ? or ?

We use SE600 as multiservice routers (LNS, Core/Edge) with last gen cards & 11.1.2.x.
It works, so far.


-- 
Olivier Benghozi
Wifirst