How to best represent "versioned" 3D data?

If I have a simulation running that manages 3D data, and where some of the
data will change after every simulation cycle, and I want to be able to keep
track of the changes, so that I could go "back in time" to a previous state,
and restart from there in case of a programming bug, how do I best
represents this in HDF5? Note that I want to protocol all changes, but in
the normal case, I am only interested in efficiently accessing the newest

One way would be to define my dataspace unlimited in 4 dimensions, and use
the index of the fourth dimension as the point-in-time where some chunk
changed, always saving a whole chunk, even if only part of it changed. Then
I would need some way, for every 3D chunk, to know all the points-in-time
where the data changed, and most importantly, what is the "highest index".
This would basically be a sparse "append-only" dataset.

Another possibility would be that I divide the 3D space in blocks of a fixed
size (that what I would normally use as a chunk size), and instead have for
every "logical application chunk" a VLEN that contains a list of composites,
where each composite contains the "point-in-time" and the new data. This
solves the "when was the data changed?" problem, but am I right to assume
that I always have to load the whole VLEN when accessing it, meaning I have
to load the "whole history" every time? Also, I only write at the end of the
VLEN, but the VLEN size has to be updated, so this usage is not
"append-only" like above.

Are there any other obvious alternatives, and why should I choose which one?

MfG / Regards,
Sebastien Diot

Grossblittersdorferstrasse 257-259
66119 Saarbruecken
Tel.: +49 (681) 8808-0
Fax.: +49 (681) 8808-300


HR A 6448 Amtsgericht Saarbrücken
Komplementär: A. Reiß & Sohn GmbH
HR B 4965 Amtsgericht Saarbrücken
Geschäftsführer: Dipl.-Kfm. Karl-Heinz Siebenpfeiffer