sql server replication: We have 2 sites which are normally connected by a 10mb fixed line, I'm
looking for a backup should this link fail. Both sites use the same software
connected to a sql server database. If the link fails both sites should be
able to continue in a limited capacity and once the link is restored the
data should synchronise.

I've been looking at this for a little while and merge replication seem to
be the right solution, however I would like to know if anyone could give me
any other options?

Thanks

Dale

you have three options

1) merge
2) transactional replication with queued updating - good where most of the
updates happen on the subscriber
3) bi-directional replication - good when your tables are naturally
partitioned.

Merge is the best solution as you can resolve conflicts. When a conflict
occured with queued you can only view the conflict, you can resolve it in
Conflict Viewer. Queued also works best when most transactions occur on the
publisher, not the subscriber.
Bi-directional replication does not have any mechanism to handle conflicts,
the jobs fail and you have to manually fix everything - similar to queued
really. Bi-directional replication is also harder to do schema changes on.

Merge and Queued add a GUID column to your underlying tables, bi-di
transactional doesn't.

[quoted text, click to view]

AND any conflicts which occur with bi-di means you have a MAJOR data cleanup
situation to deal with. You have to manually reconcile everything. That is
why bi-di is NOT recommended unless you do one of two things:
1. Enforce that transaction ONLY occur on one side or the other
2. There is a partition which ensures that no changes on one side will ever
conflict with changes on the other side

--
Mike
Principal Mentor
Solid Quality Learning
"More than just Training"
SQL Server MVP
http://www.solidqualitylearning.com
http://www.mssqlserver.com

and how so?

When you get a conflict the distribution agent stops on the side which
detects the conflict. You can then go in and manually fix it. My expireience
is that a single conflict does not mask hundereds of conflicts coming after
it, but rather is a single occurence.

The key is, as both you and I have pointed out, to minimize the chances of
conflicts occuring; but when they do; replication does fail so that you can
fix the conflict and continue.

On the bi-directional systems I have worked on, I have delivered sound
solutions which have been relatively problem free. Bi-directional
transactional is not a good fit for all requirements, but it can work.


[quoted text, click to view]

Therein lies the problem. While replication may halt, it may not also. If
you have duplicate primary keys, it will halt. If you update a nonexistent
row, update one side and delete other side, it will halt. It will NOT halt
if you update the same piece of data on both sides to different values. The
last one in will win and you will have zero indication that a previously
saved value has now been overridden which can create severe issues with the
business.

Also, replication halts in two cases. This is a very undesirable thing to
do. You can't exactly halt the transaction flow from applications, so
additional transaction begin piling up on both sides because replication is
halted. This not only dramatically increases latency, but it also has an
increasing potential to generate even more conflicts which can put you in
the same situtation with teh agents halting causing a further backup in
transactions, etc. until you hit the point where someone deems replication
to be completely nonfunctional and starts pointing fingers at the
replication engine. It has nothing to do with the replication engine, but
entirely due to the way the engine is being used and the design being
implemented.

The 2 conditions I outlined are the only way I would EVER implement a bi-di
system. I've been doing this since Sybase Rep Server 1.0 against Sybase
System 10. I've watched more than one bi-di system, 2 of them my own
designs, become complete toast because those 2 conditions were violated.
You get burned enough times, you learn not to keep sticking you hand in the
fire.

If you want to deal with conflicts being thrown in a bi-di system, that is
you choice. I certainly wouldn't want to be the one sitting in that seat,
because I know all too well just what kind of havoc a single conflict can
cause in that type of implementation. I also prefer to not have to babysit
a replication system 24 hours a day because I've placed a design that has a
flaw which can only be overcome with manual intervention and a decision by
an IT person who isn't necessarily qualified to make a call on the data.

--

Mike
Principal Mentor
Solid Quality Learning
"More than just Training"
SQL Server MVP
http://www.solidqualitylearning.com
http://www.mssqlserver.com