> How about this: Each machine maintains a list of all of it's transactions (local). It also maintains a list of all the other machine ID's that have contacted it, and last update time/date and a list of all of it's replication partners and when it received it's last update from them. At some fixed interval each machine contacts all of the partners in it's replication list and sends it it's machine ID. That machine looks up the ID, finds out the last update that was sent and then synchronizes the data. Each record in the transaction journal contains a date/time stamp so if the process is interrupted you reset the last updated time/date stamp. The receiving station will either have to roll back the changes, or make note of the changes made so that when it reconnects at some later point it doesn't try to apply the same changes it's received.
>

That is along the lines of what I'm thinking. There really shouldn't be a need for rollbacks because the most recent transaction can be considered the most appropriate.

Unfortunately, I can't use dbc (triggers). The tables must remain fp 2.6 format because C++ apps using Codebase.

There is a product for windows 95/98 called Medley from Mangosoft that guarantees transparent file based replication with redundancy across multiple machines as long as half of the nodes are online. This seemed a good solution until I found (despite advertising) that the product only works for windows 95/98. This is especially disapointing since I learned that the product was developed on NT and ported to Win 95/98. Apparently marketing considerations are preventing the release of an NT product, over the desires of the development team.

What ever we do will have to be extremely low maintenance because of travel costs/time. Uzbekistan, as an extreme example, takes 4 days of travel to get there and back. Of course on a North Altantic, North Sea, and even Gulf of Mexico rig the weather can extend your stay.

The accounting considerations are weird, it often happens that a few hundred dollars is saved on $100 thousand contract. Then there is no hesitation to spend a couple of thousand dollars in airfare alone to sent a man to correct an avoidable problem. Of course, service is billed to another account.