Is gaeo sensitive to supernode problem ? #52

Closed
Riduidel opened this Issue Feb 15, 2013 · 3 comments

Projects

None yet

1 participant

@Riduidel
Owner

Supernode problem is described (as an example) in this article : http://www.opencredo.com/blog/neo4j-super-nodes-and-indexed-relationships-part-i

Do gaedo suffer from such issues ?

Test that using Contiperf to provide effective measures.

The test should make sure that access time follows (at worse) a linear relationship with the relationship count.

@Riduidel Riduidel was assigned Feb 15, 2013
@Riduidel
Owner

If supernode is an issue for search (as it is for writing data) ... a radical solution should be found, no ? (hint : http://lists.neo4j.org/pipermail/user/2011-September/012362.html)

@Riduidel Riduidel pushed a commit that referenced this issue Feb 18, 2013
Riduidel First test for #52.
This test does not expose the bug I'm chasing right now, but I'm sure it will expose it soon. However, it exposes another one : the inability of my implementation to have fully concurrent (read non-synchronized) code being run.
5f101da
@Riduidel
Owner

What to do to lower that sensitivity ?

  • Write a test case exposing the bug (currently working on, but it seems hard to find the good test case)
  • Replace associations to most of the literals by properties on nodes. This would make "bigger" nodes, as they will no more have as only property their value, but also all the literal properties. This may not be a big change (excepted we will have to distinguish the property-on-node from the property-on-relationship, which may not be trivial).
  • Add specific property indexing and search : search on literal properties should not back-navigate the graph, but rather read a property index looking for the nodes having the correct value.
@Riduidel
Owner

So ...
To stand it clearly, yes gaedo is very sensitive to supernodes problems. And yesy this must be changed.
Hopefully, it will be changed bi fixing #68

@Riduidel Riduidel closed this Jan 27, 2014
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment