Following up on modelling approach 1)  Michael sketched previously:

With 50k trucks using a separate relationship type per truck is a bad
idea since atm Neo4j supports 65k different relationship types.

Just as a stupid idea to discuss:
* how about using 10k relationships types for all the trucks?
TRUCK_ROUTE_0001 to TRUCK_ROUTE_9999. Each individual truck is mapped
to a rel type using a modulo operation on it's identifier (think of
consistent hashing). On average one rel type is shared between 5
trucks.
* a traversal for a specific truck then need to follow just one out of
10k relationship types. Of course you still need to inspect the
relationship property to decide which of the 5 trucks it is, but it
should be faster than using the same reltype for all trucks.

/Stefan

2015-02-18 10:07 GMT+01:00 Michael Hunger <[email protected]>:
> Perhaps you can share some of your Expander code?
>
> Not really sure between what your edges are?
>
>
> Two ideas:
>
> 1) How many trucks do you have? Perhaps it makes sense to encode the
> truck-id as relationship-type? So you have fewer rels to check and can
> benefit from the separated storage by rel-type and direction.
> 2) Model the trip as a node connected to a truck, and all locations it
> visited (perhaps/optionally even encode the location-id as rel-id but that
> might be overkill) so you can quickly find all that started at "A" and then
> check if the trip has a rel to "B"
>
> 3) Another more verbose approach be to model each trip as a sequence of
> nodes (which are shadow nodes of the locations), connect the start-node of
> the trip to the truck (optionally all trip-nodes of the trip to the truck).
> And then have a relationship to each stop of the trip.
>
> I'd probably go with model #2
>
> HTH Michael
>
>
> Am 16.02.2015 um 12:54 schrieb [email protected]:
>
> I need some modelling advice.
>
> We want to store and analyse movement patterns. Think of trucks moving
> through a logistic's networks.
> We want to ask which truck has ever moved from location A to location B and
> what was the sequence of intermediate stops they made to get there.
>
> In a later stage we also want to be able to ask this question if there is no
> truck that has stopped at location A and B. Which trucks and which sequence
> of stops would we have needed to get from A to B.
>
> Right now we modeled all locations as nodes and every trip a truck has ever
> made as a separate edge. The edges are attributed with a truck ID and a
> sequence number.
> We wrote our custom expander class to be used with the traversal framework
> and to take care of the sequence numbers and truck IDs to only get complete
> sequences for individual trucks.
>
> However, this performs very badly.
> Right now we have 300 locations/nodes and 300.000 trips/edges. Some stops
> have 20.000 outgoing trips that we are checking for truck ID and sequence
> number (for every outgoing relationship, get attributes and check) .
> This performs too badly. 13 seconds for 900 sequences.
>
> Finally, we want to try to scale it to 3000 locations and 20.000.000 trips.
>
>
> Do you have any alternative modelling ideas?
>
> Thanks a lot already.
>
>
> ps: I was thinking of storing every trucks list  as a long linear sequence
> of stops/nodes. The nodes are additionally linked to some identifier Node
> through a type of is relation: "stop x is location A".
>
> --
> 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.

-- 
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.

Reply via email to