I'm involved in a project, for PDAs, that consists of upgrading our linear database. We basicly use text files for our data. I wanted to create a new DB, based on an AVL tree, to really increase our search times, and upgrade our file format from text to binary. This means I would have to serialize the AVL tree to file. Problem arises when we become aware that this is a critical database software, that should be practicly crash-proof. When the DB was text-based, changes revolved around adding another line to the file, but with a binary file, a small change "here" can change the whole file, and one of the things about such databases, is that any commits should be as fast as possible, because that minimizes the risk of data corruption in the event of a crash. The only solution I've came up with is to create a "bloated" file so that I have empty areas to write into. This would mean that adding new data to an area of the file wouldnt affect the rest of the file. Any thoughts you guys might have on all of this would be apreciated.