A 100 friends per customer is a somewhat high cardinality. Going out four levels you get 100 to the power of 4, 100.000.000 friends out of 10.000, then "distinctified". There's probably a good chance that the result is just the complete Customer set. You should take a look at the query plans for both and see if MySQL is able to come up with a smarter plan. Also make sure that you have indexes on the initial customer lookup. And the queries have to be run many times to let each database warm up properly. Neo4j needs more warmup iterations because the database itself is JIT compiled by the JVM. If you reduce the degree, say to 10 for instance, you might see Neo4j perform better than MySQL; especially if you also increase the search depth.
-- Chris Vest System Engineer, Neo Technology > On 30 Jan 2017, at 19.02, Artur Rudalski <[email protected]> wrote: > > Hi all! > > starts from beginning, I would like to check performance MySQL and Neo4j in > graph traversing therefore I created sql database composed of two tables: > customer and customerfriend. Second one contains two keys: CustomerID, and > FriendID which point to CustomerID in customer table. In Neo4j I created > corresponding entities - 'Customer' nodes and 'FRIEND_OF' relations - > (c:Customer)-[f:FRIEND_OF]->(cc.Customer) . Both databases contain same data: > 100000 Customers, each Customer has 100 relations. > I perfomed tests using below queries: > > Neo4j: > match (c:Customer{CustomerID:99})-[:FRIEND_OF*4]->(cc:Customer) (240s) > return distinct cc.CustomerID; > > MySQL: > > SELECT distinct cf4.FriendID FROM customerfriend cf1 (60s) > join customerfriend cf2 on cf1.FriendID = cf2.CustomerID > join customerfriend cf3 on cf2.FriendID = cf3.CustomerID > join customerfriend cf4 on cf3.FriendID = cf4.CustomerID > where cf1.CustomerID =99; > > I have tested queries for 50 relations per customer, and I achieved same > results (Neo4j slower than MySQL) > > Queries are run from simple Java application, which just connect to database > (using available connectors), run queries and measure execution times. > > Finally I recieved unexpected results- Neo4j is slower than MySQL in > performing above query (Neo4j 240s, MySQL 60s). Before starting my tests I > read some articles which shows that Neo4j works better in case of recursive > queries than MySQL, that's why I'm start wondering if I'm doing something > wrong or measured times are proper (?). > I'm wondering if in Neo4j exists any possibilites to tune system performance > (increase buffer size or something). In case of MySQL I set up > innodb_buffer_pool_size to 3G which affected better query performace (shorter > execution time). > > These tests are crucial for experimental part in my myster thesis and I'll be > very greatful for any hints and help > > Thanks in advance! > > Artur > > -- > 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. -- 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.
