KCC topology and replication Q's

[Diane McCorkle]

we've implemented the following

5 sites in a "High Speed" site link with a cost of 50 that we can refer to
as group "A"
2 Sites in a "Medium Speed" site link with a cost of 75 that we can refer to
as group "B"
14 Sites in a "low speed" site link with a cost of 100 that we can refer to
as group "C"
We've cleared out any previous information the KCC had automatically
generated and are letting it regenerate.

There have been a few discussions regarding the following scenarios:
Should we assume that Group C sites should be trying to replicate to and
from group A sites?
What we hope does not happen is the sites in group C all try to replicate
with each other and never go up to A.

What was also brought up is a question that on some occasions the low
speed sites are in a situation where they can not ping another low speed
site, the assumption has been if they can not ping, then KCC should fail
them and not consider them valid replication partners and move on to find an
available partner in the hi or med speed groups,

The counter to that is an opinion that when KCC sees a site in AD S&S, it
tries to ping it, when it fails it logs an error, and isn't smart enough to
just ignore it and move on

A clarification on the two above opinions of KCC and ping would help us a
lot

Thanks! Diane

 

[Jody Flett]

Assuming all of the DC's are in the same Domain and therefore replicate the
same partitions....

Site link bridging is enabled by default so you can assume that all sites
can talk to each other.

All of the sites that you have put into a single site link assume that they
can all talk to each other, and also there is no cost to replicate with each
other, they are using the same site link after all. The sites within the
site link will make a hub spoke topology with each other. The trouble with
this being that you cannot nominate who the hub is ... the KCC picks this...

This hub server will probably have connection objects out to the other
sites, rather than all of the servers in the site link making connection
objects to other sites, I think.

Of course if DC's are GC's and you have multiple domains they need to make
the relevant connection objects in order for them to get all of the
information from other domains etc... if this information resides in other
sites then it will have to make connection objects to the other sites.

On the second point if a site becomes unavailable, the KCC will attempt to
make connection to this site for a period of time, but if it can get the
information it needs from anther dc in another site it will recalculate the
link and tombstone the original link whilst this happens it will put errors
into the event log, and i think that the KCC will continue to complain that
it cannot make a connection object to the unavailable server until the
administrator intervenes (it runs the same algorithm each time therefore it
will always come up with the same result). However if the KCC cannot make a
link to get the information that the DC needs you will get a different error
stating that there is not enough connectivity to replicate whatever
partition is failing.

I think that to achieve what you want to do, you really need to set up your
site links to reflect the physical topology of your network. Ie an AD site
link, individually costed for each physical site rather than putting all
sites that you want to have the same cost into the same site link, as this
will not create the topology that you are after.....

I hope this all makes sense, I have read through it a few times and it seems
to but please post back if you have any queries....

HTH

Jody