1. Janusgraph-Dev
  2. Topics

What's JanusGraph's strategy to reduce instance-wise communication in clusters?

i...@...
 

When querying against a large graph, great time costs may occur. For example, if a vertex's adjacent vertices lies on another machine, then queries like "find father of this person" may slow down due to network communication.

Is there any strategy or optimization JanusGraph uses to reduce such costs? Considering that SimpleBulkPlacementStrategy simply assign new vertices randomly to different partitions, which means adjacent vertices are likely to be in different machines.
ankur...@...
 

Explicit graph partitioning can used for this but for large graph: http://docs.janusgraph.org/latest/graph-partitioning.html .


On Monday, December 18, 2017 at 2:09:53 AM UTC+5:30, i...@... wrote:
When querying against a large graph, great time costs may occur. For example, if a vertex's adjacent vertices lies on another machine, then queries like "find father of this person" may slow down due to network communication.

Is there any strategy or optimization JanusGraph uses to reduce such costs? Considering that SimpleBulkPlacementStrategy simply assign new vertices randomly to different partitions, which means adjacent vertices are likely to be in different machines.
Calvin Lei <ckp...@...>
 

The documentation needs to be updated. The config cluster.partition = true is no longer used as it is commented out in the source.


On Sunday, December 17, 2017 at 9:03:41 PM UTC-8, an...@... wrote:
Explicit graph partitioning can used for this but for large graph: http://docs.janusgraph.org/latest/graph-partitioning.html .


On Monday, December 18, 2017 at 2:09:53 AM UTC+5:30, i...@... wrote:
When querying against a large graph, great time costs may occur. For example, if a vertex's adjacent vertices lies on another machine, then queries like "find father of this person" may slow down due to network communication.

Is there any strategy or optimization JanusGraph uses to reduce such costs? Considering that SimpleBulkPlacementStrategy simply assign new vertices randomly to different partitions, which means adjacent vertices are likely to be in different machines.

1 - 3 of 3