How many graphs can i create in one DGraph installation on cloud service

Users
#1

I have a situation where each user will have a million node graph of their own which possibly connects to other users’ graphs. I want to implement graph belonging to each user as separate graph in DGraph. How many such individual graphs can DGraph installation can allow if the users are in billions? I have this type of usage because work on individual user’s graph is far more than work on graph connecting multiple users.

Also, what is the Big O for finding cycles in the graph? What algorithm is used?

Ram

#2

Hi @ramk,

I’d recommend having just one instance of Dgraph and putting all your data there. Dgraph’s performance shouldn’t decrease as a result of clustering effect. But, if you can explain the kind of queries you want to run, I can give a better idea about whether this cluster effect would affect query performance.

We currently don’t detect cycles in graphs. That’s something planned for later.

#3

Thanks Manish! If it performs well for what i want then that is a good enough way forward.

I’m going to have to find possibility of loop before connecting any node to any node with a uni-directed edge. The graph is going to be very complex, though there is simplicity in edge type(single type). The only query i will have to do at present is for getting all the paths a node is on.

Future queries will be about what nodes a subgraph(belonging to a user) is connected to.

One thing why i thought about maintaining separate graph for each user is that i can spawn threads to make queries on multiple users and determine what i want in parallel.

Another question i have is, is there a maximal size or advantageous size to storage attached to each instance of compute in a cluster running DGraph?

#4

Are you saying there’s only one predicate? If that’s the case, then all your data would lie on a single machine.

You can do that even now with Dgraph. It supports concurrent execution and using more threads on a multi-core server should give you better throughput, up to a certain point.

Not really, no.

#5

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.