I have a scenario in which we have a legacy application running on SQL 2000 and want to move some scripting/rules which are run against data to another server which is running SQL 2008. We want to setup a replication scenario so that updates made to specific tables/columns in the SQL 2000 system are synchronized with the data on the 2008 rules server.
The SQL 2008 system is also using SSD's, so we are trying to limit the number of writes.
My first question: From my understanding Merge Replication needs to be setup going from the 2008 to the 2000, does this hold true with Transactional? Or can the Transactional go directly from 2000 to 2008?
The 2000 SQL database despite being a legacy application, still gets a lot of updates (we are working to phase it out, but it won't be for a while still). So the two replication methods we can use become limited (again based upon my understanding of how replication is intended) to Transactional or Merge.
We don't require up to the moment updates as these rules are always run against previous days (normally previous weeks) data - however, there is some concern over adding additional columns to the DB's due to our still using the legacy application which accesses it. The updates also only need to go one direction, we only need the 2000 to update the 2008 - the 2008 (via replication anyways) will not be performing any updates.
My section question: This leads me to believe that our best option may be a Transactional Replication. I wanted to get others feedback to find out if they believe this is the best method to use.
My last two questions: The next two questions I will be working on testing in a development environment, but wanted to check if they were even possible before I tried troubleshooting them (this is also assuming that, like merge replication, transactional replication requires 2008 to be the publisher which was the first question above):
1) We are only planning on replicating the columns needed, which means that our 2008 server's tables will have fewer columns than our 2000 - but the 2000 is actually the subscriber. Is it possible to replicate to a table with more columns than the source table?
2) Can we "begin" the transactional replication with the 2000 database being already populated? The issue of course being that the 2000 is actually the true source - so what we are going to do is do an initial DTS to the 2008 then setup the transactional replication.