Distributing DNS servers

Barrett Richardson barrett at phoenix.aye.net
Wed Sep 1 04:32:37 UTC 1999



On 31 Aug 1999, Bill Larson wrote:

> I've been following this thread about DNS server failover capability
> and have been quite impressed.
> 
> One question though, how quickly can OSPF or BGP routes be switched,
> and what triggers it?  Are you talking about an automatic detection
> process that triggers the routing table changes, or is it performed
> manually?  And, does the routing changes occur in seconds or minutes?
>

If a DNS box were participating in OSPF, other OSPF participants
would delete routes that is has broadcast after not hearing a
HELLO for a configured RouterDeadInterval (default is typically
40 seconds). If gated is able to do an Link State Advertisement
informing OSPF participants something to the effect that it's
links are being downed (when it gets a SIGTERM), it could be much
quicker -- I've yet to test this. Worst case is the RouterDeadInterval.
If a DNS box were peering with routers with BGP, it would send
a CLOSE to its peers on receipt of a SIGTERM. In that case
route removal would be immediate.

Another case is the event of network problems and a DNS box
participating in a routing protocol can't get information to
neighbors/peers. OSPF would rely on the RouterDeadInterval,
BGP would rely on a holdtimer. The RouterDeadInterval must
agree with the rest of the OSPF speakers in an area, but the
holdtime can be configured on a per peer basis in BGP. This
makes BGP more flexible in that regard. The holdtime by
default is typically 180 seconds, but with a DNS box on
an attached segment where minimal routing information is
exchanged, a small value would probably be favorable. If
a BGP peer doesn't hear a keepalive (or update) in holdtime
seconds, it drops the connection and deletes routes learned
from the downed peer.


-

Barrett

> Thanks,
> 
> Bill Larson (wllarso at swcp.com)
> 
> 
> 



More information about the bind-users mailing list