No doubt it could be done, Jim - but then the traversals get more complex of course. Ideally it would largely transparent via the index framework. Alternatively, I wonder if there is work that could be done at the kernel level to deal with optimizing frequent relationship attachment/detachment on "hotspot" situations/supernodes.
-----Original Message----- From: user-boun...@lists.neo4j.org [mailto:user-boun...@lists.neo4j.org] On Behalf Of Jim Webber Sent: Monday, May 02, 2011 8:48 AM To: Neo4j user discussions Subject: Re: [Neo4j] Lucene/Neo Indexing Question I actually think this discussion on supernodes is very valuable - we've seen it crop up in a small, bit significant, number of implementations and we've seen credible and hairy workarounds. [pure speculation follows] I'm wondering whether it would be possible to intercept the write to the "red" colour node (as in Rick's domain) and redirect that to an underlying balanced tree (effectively an index of red things). That is, the "red" node is actually a function which as a side-effect attaches relationships to nodes in a balanced tree-of-red. All of which happens without any explicit stopping and calling out to separate indexes. I suspect this is non-trivial given we optimise around a stable, performant on-disk structure, but I'd love* to hear the kernel hacker's views on this. Jim * Unless those views are that I'm a bozo, then I'll just reluctantly hear them for the sake of completeness. _______________________________________________ Neo4j mailing list User@lists.neo4j.org https://lists.neo4j.org/mailman/listinfo/user _______________________________________________ Neo4j mailing list User@lists.neo4j.org https://lists.neo4j.org/mailman/listinfo/user