Another thing to consider is that your traversal is loading node properties, which can slow it down when compared to a traversal based only on relationship types. However, having said that, your graph is quite small, and so the entire graph should fit in memory with the right cache settings, and then hitting the node properties should not make much difference.
If you are likely to scale this to much, much larger graphs, you might want to consider restructuring so your traversals do not need to test for the node type. This is can be possible by implying the node type from the relationship types. You might need to use more relationship types to do that. On Wed, Feb 23, 2011 at 10:08 AM, John Nonak <[email protected]> wrote: > Hi again, > > Well well well, yep the cache did help, even if the queries were different > ones > :) It's obviously not just caching the final list of paths, as I naively > assumed. > > I got the requests down to about 150ms which looks much better. I suppose > with > proper optimization I could even get better results. > > > Thanks again > Regards > > John > > > > ________________________________ > From: Peter Neubauer <[email protected]> > To: Neo4j user discussions <[email protected]> > Sent: Wed, February 23, 2011 10:25:43 AM > Subject: Re: [Neo4j] Noob question about performance > > John, > how many nodes and relationships do you have in your graph? Are you > running only once without warming up the caches (you might want to run > the calculations say 10 times and take the mean)? > > Also, it depends on how many nodes you are "touching" during your > traversals. The number of hops needed to complete a traversal is the > defining measure for the total time in Neo4j. Maybe you could keep a > counter in your graph algo and print it out? > > Cheers, > > /peter neubauer > > GTalk: neubauer.peter > Skype peter.neubauer > Phone +46 704 106975 > LinkedIn http://www.linkedin.com/in/neubauer > Twitter http://twitter.com/peterneubauer > > http://www.neo4j.org - Your high performance graph database. > http://startupbootcamp.org/ - Öresund - Innovation happens HERE. > http://www.thoughtmade.com - Scandinavia's coolest Bring-a-Thing party. > > > > On Wed, Feb 23, 2011 at 9:21 AM, John Nonak <[email protected]> wrote: > > Hello, > > > > I am starting to evaluate neo4j and I wrote some basic test application. > The > > results I get are not as fast as I would have wished :) but of course it > is > > probably because it is not optimized or simply because my case do not fit > that > > well with graph DB. > > > > If somebody would have few minutes just to tell me if neo4j is a good > choice > >for > > our application, here the story (sorry for the terminology I use, I m new > to > > graph DB in general) > > > > - Two types of nodes, A and B > > - The graph always alternate A and B nodes, path example : A -> B <- A > -> B > > (never A -> A -> B) > > - Maximum depth of a path is 4, like in the previous example > > - There are two different types of relationships > > > > - The application needs to find all the paths between a given A node and > a > >given > > B node, where the distance between the two nodes is at maximum, and where > the > > two first relationships are identical. So for instance I want to find all > the > > paths between A1 and B5 with the following graph: > > > > A1 --Ra--> B1 <--Ra-- A2 --Ra--> B2 > > A1 --Ra--> B1 <--Ra-- A2 --Ra--> B5 > > A1 --Ra--> B2 <--Rb-- A3 --Ra--> B5 > > A1 --Rb--> B3 <--Rb-- A4 --Rb--> B4 > > A1 --Rb--> B3 <--Rb-- A4 --Ra--> B5 > > A1 --Rb--> B3 <--Rb-- A5 --Rb--> B5 > > > > So the end result would be > > A1 --Ra--> B1 <--Ra-- A2 --Ra--> B5 (Same Ra relationships linking A1 and > A2 > to > > B1) > > A1 --Rb--> B3 <--Rb-- A4 --Ra--> B5 (Same Rb relationships linking A1 an > A4 to > > B3) > > A1 --Rb--> B3 <--Rb-- A5 --Rb--> B5 (Same Rb relationships linking A1 an > A5 to > > B3) > > > > The point is to compute the amount of final relationship types between a > node > A > > and B. So in this case we have 2 Ra types and 1 Rb type between A1 and > B5. > > > > Hope this is clear enough :) > > > > > > I run my test with the following data: > > - 1000 A nodes > > - 100 B nodes > > - Random linking between A and B nodes. Between 20 and 50 relations ships > for > > each single A node to B nodes. > > > > I made two different implementations: > > > > 1) Extend the AllPaths class from the graphalgo lib and modify the filter > to > > analyze the two first relationships > > 2) Similar to AllPaths implementation but with the use of an Evaluator to > > analyze the two first relationships but also exclude wrong path faster > > > > Once I get the paths I iterate them to compute the total of final Ra and > Rb > > relationships. > > > > The result for all paths between two specifics nodes A and B returns 2573 > > different paths > > > > Implementation (1) processes it in about 1150-1250ms > > Implementation (2) processes it in about 950~1050ms > > > > I tried to tweak the mapped memory config, but so far this is as good as > I > > manage to get it. I am using the embedded java version of neo4j. > > > > I believe that the more nodes and relationships I will add the more time > it > >will > > take to compute. So the big questions are: > > > > - Does this scenario makes any sense under neo4j/GraphDB? > > - Are the performances that I am getting normal considering the case or > is > >there > > something I am missing or doing wrong? > > > > > > Thank you for your time. > > Regards > > > > Nonak > > > > > > > > _______________________________________________ > > Neo4j mailing list > > [email protected] > > https://lists.neo4j.org/mailman/listinfo/user > > > _______________________________________________ > Neo4j mailing list > [email protected] > https://lists.neo4j.org/mailman/listinfo/user > > > > > _______________________________________________ > Neo4j mailing list > [email protected] > https://lists.neo4j.org/mailman/listinfo/user > _______________________________________________ Neo4j mailing list [email protected] https://lists.neo4j.org/mailman/listinfo/user
