Hi there,

Not sure this might help you out but in a project I'm working on, we have decided to go with an MD5 hash that is computed via a script task in our data flow (there are a bunch of ways to do this, that's the one we went with because we could easily switch for another scheme). The hash is saved in the dimension. When we get rows from the operational system, we compare the hash and based on that decide whether an update/insert/no action is needed.

The thing is that with checksums, hashes etc. you can have collisions, you just have to check how high the risk and if you can ever afford one. One could argue that it's the same with GUIDs, while it's highly improbable, it's not impossible. If you can't afford that then you're better off with either a column-by-column comparison or an extra column in your source data that can tell you the latest time something was changed or something like that.

Hope this helps,

Greg