DistGraph: Get local and remote/halo nodes in each partition

Questions
#1

I can get the total number of nodes in a partition from the json file created by the partition_grapy.py in the dist graphsage example.

How can I get the following information for each partition from DistGraph object?

  • Number of Local Nodes
  • Number of Remote Nodes
  • Which Remote Nodes belong to which partition?

Consider the following very simple example:

Number of Partitions: 3
Total nodes in partition 1: 0,1,2,3,4,5,6,7,8,9
Local nodes in partition 1: 0,1,2,3,4
Remote nodes in partition 1: 5,6,7,8,9

For partition 1:
Number of Local nodes: 5
Number of Remote nodes: 5
Nodes 5,6,7 belongs to partition 2
Nodes 8,9 belongs to partition 3

How to get this information from DistGraph?
How to get the feature vector or a node or multiple nodes from DistGraph?

#2

@Rhett-Ying @zhengda1936 Can you assist please?

#3

g.ndata['inner_node'] == True indicates local nodes. False Indicates remote nodes.

As for remote id to partition id, please refer to dgl.distributed — DGL 1.1.1 documentation. Here’s an example in test: https://github.com/dmlc/dgl/blob/a52be5c683cd491ee5f81d38670472017eecc481/tests/distributed/test_partition.py#L165-L169

#4

@Rhett-Ying Thank you for your assistance.

Does g.ndata['inner_node'] == True gives a list/tensor of the local node id’s?

#5

yes, the true masks indicates local node IDs

#6

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