Hi John, Gotcha better now.
[snip] > I am not sure I understand. Why do large graphs that span multiple machines > require domain specific sharding? Assuming a not overly inter-connected set > of shards, can't you have an intelligent query processor that partitions a > query (passes on a subquery to the next server) whenever a "off server > reference" is encountered at query run time? If you've got a sparsely connected graph, then yes you can do this. Neo4j doesn't do this at the moment however. It'd be up to you application to notice special keys/values in your nodes, and then make a second call over to the other server. > My graphs are large and distributed so replicating each shard into the > master won't work. Writes are only done within a single shard so only the > references between shards would need to be updated. I need to wrap existing > data stores. Understood - so as a product catalog changes with reference to its sales blurb (different instances) you'd want to keep them in sync. Right now that consistency would fall to your application to manage. BTW - how big are your graphs that they can't fit inside a large, replicated cluster? This might make a really interesting use case to drive out some of our future requirements if you can share. Jim _______________________________________________ Neo4j mailing list User@lists.neo4j.org https://lists.neo4j.org/mailman/listinfo/user