[prev in list] [next in list] [prev in thread] [next in thread] 

List:       isp-bgp
Subject:    [isp-bgp] Re: iBGP not working properly ?
From:       Erik Haagsman <erik () we-dare ! net>
Date:       2003-05-22 22:48:33
[Download RAW message or body]

Hi Wim,

> However, when the path length is the same and we perform a 
> traceroute on both Router1 and Router2 to the same host, they 
> both go out via a different link... their own. I don't think 
> that's normal, right ?

Yes, in case of equal path length this is perfectly normal since each
router will prefer a direct route (it's own connection to the transit
peer) over a route with an extra hop (the neighbor iBGP router). If you
want to solve this, use a form of route-map or configure the iBGP
neighbors' weight to prefer one over the other. Since you're connected
with two different transit ASN's, bgp route-sharing isn't an option, so
you won't be able to load-balance perfectly over the links, but with
some creative use of route-maps and weighting you'll come a long way.


> If I read this correctly, Router1 only sent 20896 prefixes to 
> Router2, although it received 120248 from Transit1. On the 
> other hand, Router2 sent 115223 to Router1 and it received 
> 120464 from Transit2... where does that difference come from ? 
> Why doesn't Router1 send all of its routes to Router2 ? Is 
> that the reason why most routes go via Transit2 ?

What could be your problem is that Router2 recieves more best-routes to
Transit2 than Router1 does from Transit1.
By default, only the best routes are exchanged, so if Router2 has the
most best routes, and the weighting of Router1 and Router2 are equal,
ROuter1 will accept more routes from ROuter2 than vice versa. Again, if
you want to solve this, you'll have to manually start prioritizing your
iBGP peers through route-maps/weighting. Depedning on the type of
equipment, there is usuallu also a parameter that allows you to exchange
all routes, even those that aren't installed as best route. By setting
this, both rouers would exchange a full BGP feed both-ways, but you'd
normally want optimal routing, so there's not much point in this.

Hope this helps, if I made some wrong assumptions or if I'm being very
confusing (it's happened before) let me know.

Cheers,

-- 
Erik Haagsman
Network Architect
We Dare BV
tel: +31(0)10-7507008
fax: +31(0)10-7507005
http://www.we-dare.nl


> 
> 
> Any suggestions would be very nice.
> 
> 
> Kind greetings,
> 
> Wim
> 
> ------------------------ANNOUNCEMENT---------------------------------
> ---------------------------------------------------------------------
>                          >> ISP Technology <<
> VPN, DNS Security Management, spam, 3G, WLAN Security, P2P, VoIP, PLC,
>                     802.11i, 802.1x and much more!
>        http://www.wispcon.com/spring2003/attend-sessionlist.asp
>         The 10th Annual ISPcon, Baltimore Waterfront Marriott
>                              April 23-25
> ----------------------------------------------------------------------
> ----------------------------------------------------------------------
> 
> _______________  The ISP-BGP Discussion List  ______________
> To Join: mailto:join-isp-bgp@isp-bgp.com
> To Remove: mailto:remove-isp-bgp@isp-bgp.com
> Archives: http://isp-lists.isp-planet.com/isp-bgp/archives/



------------------------ANNOUNCEMENT---------------------------------
---------------------------------------------------------------------
                     >> Recycle Your Hardware <<
              Clean out your closets and make some cash.
               Reach thousands of ISP equipment buyers.
                     http://www.isp-equipment.com
----------------------------------------------------------------------
----------------------------------------------------------------------

_______________  The ISP-BGP Discussion List  ______________
To Join: mailto:join-isp-bgp@isp-bgp.com
To Remove: mailto:remove-isp-bgp@isp-bgp.com
Archives: http://isp-lists.isp-planet.com/isp-bgp/archives/
[prev in list] [next in list] [prev in thread] [next in thread]