Hi, The last configuration for that store looks better. If you are doing traversals and accessing properties for each hop on node/relationship you need to increase the memory for the property store. If you are requesting string properties you have to increase the size some there too.
This may be a better configuration in such a scenario: neostore.nodestore.db.mapped_memory=500M neostore.relationshipstore.db.mapped_memory=10G neostore.propertystore.db.mapped_memory=10G neostore.propertystore.db.strings.mapped_memory=10G neostore.propertystore.db.arrays.mapped_memory=0M Remember the mapped memory is allocated outside the heap so you have to withdraw that from total memory when calculating what heap size to use. I would suggest to start the JVM with following parameters: java -d64 -server -XX:+UseConcMarkSweepGC -Xmx20G To speedup a cold start you should make sure the OS has cached as much as possible of the store files. An easy way to do that on unix is to just read the files sequentially multiple times. For example: dd if=neostore.relationshipstore.db of=/dev/null bs=100M Finally, what type of traversal/walk over the graph are you doing? For example, finding a single shortest path depth < 4 on a equal size graph (random distribution of connections) running on that machine should execute in 10-20ms. Regards, -Johan 2010/2/24 Miguel Ángel Águila <[email protected]>: > The configuration of the machines is a HP Proliant ML370 G5: > > - 2x Quad Core Intel® Xeon® E5440 2,83 GHz > - 64Gb de memòria > - 2Tb disc (SFF) > > And I've never finished the walk, the machine always crashes and I have > to restart, therefore I can't see the error. I think that the walk could > finish in 4 or 5 days. > > > > El mié, 24-02-2010 a las 08:21 -0400, Gutemberg Vieira escribió: >> Also, could you post the time to complete the walk? _______________________________________________ Neo mailing list [email protected] https://lists.neo4j.org/mailman/listinfo/user
