From: Peter Van Oene (pvo@xxxxxxxxxxxx)
Date: Thu Apr 05 2001 - 12:34:15 GMT-3
I'm not sure about this though I don't work that much with Cisco routers. I wo
uld suggest that ensuring reachability to EBGP Next_Hop addresses is pretty muc
h fundamental to designing BGP networks and pretty rather to EGP/IGP synchroniz
ation.
As far as how routes get into the BGP table, I'm not sure I agree with you. It
is my understanding that all BGP paths received from a neighbor will populate
the BGP RIB-IN table irrespective of whether they meet the criteria for re adve
rtisement or use. With synch enabled, the NLRI information (prefixes) must matc
h IGP provided routes in the routing table for those prefixes to be posted in t
he routing table. Reachability to the Next_Hop is also a criteria at this poin
t. Whether synch is enabled or disabled however, that path information will c
ertainly populate the RIB-IN.
As far as synch being beautiful, I am somewhat confused. I assume you mean tha
t because you have IGP reachability for a prefix, if that prefix happens to not
be posted via BGP due to Next_Hop reachability not being met, that the router
will still forward packets based on the IGP? I would suggest that this is a co
rner case benefit that does more harm than good. Its like saying a default rou
te will enable packets to flow even if you mess up your routing configuration.
In this case, I would think it far better for things not to work so that the d
esigner could immediately notice and resolve the configuration issue. Cludged
networks that work cause more grief than ones that don't as the need to troub
leshoot is much more obvious :)
Keep in mind that the networks that sync was designed to support no longer exis
t. Those networks provided transit services to an internet with so few prefixe
s that the IGP could handle them. I just wanted to reiterate the sync is obsol
ete and that one shouldn't waste ones time trying to figure out how to make it
work :)
Pete
*********** REPLY SEPARATOR ***********
On 4/5/2001 at 9:51 AM Mark Salmon wrote:
>I have a caveat to the sync question. I would make sure that ALL IP nets
>(include the WAN/LAN EBGP) links are reachable by all BGP routers in your
>AS. The beauty about sync is no IP subnets/CIDR nets will appear in the
>BGP table unless they appear first in the IGP table. With Sync off (no
>sync) that will not happen. If your design is not done properly (ie all
>IP nets in the BGP table include next hop) is not reachable, then packets
>will be dropped.
>
>Peter Van Oene wrote:
>
>> Inline comments
>>
>> *********** REPLY SEPARATOR ***********
>>
>> On 4/5/2001 at 10:49 AM Oscar Diaz Poveda wrote:
>>
>> >Dear all,
>> >
>> >Inside an AS where there are routers that are not running IBPG:
>>
>> Obviously a stub AS as opposed to a transit? If transit, you should run
>IBGP. If full routes internet transit there is no other option here
>(assuming scalability tools like rr's/confeds etc used as well)
>>
>> >
>> >When you should redistribute BGP into the IGP and when you should turn
>off
>> >synchronization???
>>
>> Synch is only relevant when you run a transit AS and don't' fully mesh
>with IBGP. However, you should NOT do this. This is bad. Very bad in
>fact. If you are running a transit service, use IBGP and disable synch.
>Synch is an obsolete feature that should never be turned on. In fact, I
>highly doubt that any transit AS in the world has synch enabled. Further,
>I suggest that since it has received little to no programming attention
>(educated guess) I expect it doesn't even work flawlessly. If a proctor
>asks you to enable synch, I'd call an exorcist immediately.
>>
>> For what its worth, Juniper (who make routers ostensibly for transit
>as's) do not even have a knob to enable synch.
>>
>> Pete
>>
>> >Thank you for your advice in advance.
>> >
>> > Best regards,
>> >
>> >Oscar.
>> >
>> >
>>
This archive was generated by hypermail 2.1.4 : Thu Jun 13 2002 - 10:29:41 GMT-3