Hello Mark,

On Sun, 2011-01-16 at 14:21 +1100, Mark James wrote:

> Is there an easy way to determine the amount of disk space an
> individual quad store graph is taking up?


Indexes for IRIs and literals are common for all graphs, so it's not
obvious how to count the consumed space if an IRI is used in more than
one graph.
Accounting for quads becomes unclear too.
Before version 6 it was possible at least to get the size of indexes for
quads, divide it by total number of quads in all graphs (to get a disk
cost of a single quad) and multiply by number of quads in the specified
graph, thus some adequate number was available. Starting from version 6,
there are partial indexes on S,P and O,P, as a part of "3+2" indexing
model, and data of these indexes are shared for all graphs, so not
counting is possible for same reason as for IRIs. In addition, indexes
become bitmaps with compression that can sometimes save significant
amounts of additional data "for free".
So I don't know what to recommend as a reasonable measurement method.
Existing real-life storages may give some hint, but not more,
experiments with synthetic data like benchmarks does not give even a
hint, because they're "too synthetic".

Best Regards,

Ivan Mikhailov
OpenLink Software
http://virtuoso.openlinksw.com



Reply via email to