Hi, Mark,
sorry for the late response: 1) How much data is there on each node/edge? Each node has only several string attributes. Therefore the node size is up to 200 bytes. 2) What was the memory setup for both DBs? As Neo wants to load everything into memory your 1.2 million nodes and 3 million relationships will put a strain on your 4Gb box. Due to the small size of nodes, we do not need to handle with memory setup. 3) Can we see the Neo configuration? The Neo configuration was the default settings. The results was sufficently good (compared to Postgre), that we have decided not to play with configuration during benchmarks. Tunning was planned after finish of part of the implementation. 4) Why do you say Neo was bigger on disk (looks smaller in table)? Yups, that was my mistake, I accidentally switched the tables. It is ok now. 5) What protocol were you using to talk to Neo? I do not know if I understand the question. We have used both cypher and direct Java api, the Neo4J was embeded in the application. Thank you! Dne čtvrtek, 4. září 2014 13:01:34 UTC+2 Mark Findlater napsal(a): > > Hey Jan, this is really great I'll have a look at it properly later. A > couple of quick questions: > > 1) How much data is there on each node/edge? > 2) What was the memory setup for both DBs? As Neo wants to load everything > into memory your 1.2 million nodes and 3 million relationships will put a > strain on your 4Gb box. > 3) Can we see the Neo configuration? > 4) Why do you say Neo was bigger on disk (looks smaller in table)? > 5) What protocol were you using to talk to Neo? > > I guess it is "obvious" that using the API will always be a little faster > than using the interpreted CYPHER statements (because of the nature of the > product), but maybe you could share what the Cypher query was and what you > replaced it with? > > M > > On Thursday, 4 September 2014 11:34:35 UTC+1, Jan Andrs wrote: >> >> We needed to decide whether is better to create metadata repository in >> our client's data warehouse with Neo4j or with PostgreSQL, so we ran a >> little benchmark test: >> >> http://mantatools.com/metadata-repository-benchmark-postgresql-vs-neo4j/ >> >> Major reason was to decide if is better to to use graph databases or >> traditional relational ones. Neo4j performed well, but not with Cypher, we >> had to do our custom solution using Neo4j Java API. >> >> Do you have any experience with similar tests? We are also planning to >> run another one with more databases and with more conclusive results. >> >> >> -- You received this message because you are subscribed to the Google Groups "Neo4j" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
