Yes, it's like this. Each field is individually sized and only takes up the appropriate ammount of storage on disk. This is not the case with the earlier versions of NDB tables. They were always fixed width but that has recently changed. Besides, I do not think you were asking about NDB's in-memory allocation scheme, were you?

This is an example of neither fixed-width or variable storage as you are asking the database to resize a storage location based on the largest length added to the database. It would be quite inefficient to program a database this way. Imagine if you had 10000 rows of data setup like this and added a new row where the text field was 400 characters long, you would have to resize all 10000 previous rows to match the new value. It's easier to either let each field have its own size (as above) or completely fixed throughout. The advantage of being completely fixed is in lookup speed. You always know exactly how far to move in memory/disk to get to a particular piece of data. The downside to completely fixed field sizes is that it wastes so much space.

Good question!