depending on the availability of the server, you may be able to ship the last log, etc. if the server is completely unavailable, there's not much you can do except accept that some data will be lost. what we tried to do was to gauge how much time we could afford to lose in the event of a system failure. obviously you can configure longer or shorter time intervals, but the less time you can afford to lose, the more you need to look at something like clustering or perhaps database mirroring in SQL2K5 that provides more real-time synchronization.