Hi,
Here are two tickets that will allow us to gut most of the MapReduce jobs from
TraversalVertexProgram.
https://issues.apache.org/jira/browse/TINKERPOP-1166
https://issues.apache.org/jira/browse/TINKERPOP-1164
What do we gain by doing this?
1. You can have mid-barrier traversals!
2. You don't need to do a second scan of the graph in a MapReduce phase
to compute the sideEffects.
The two notes above are functions of the fact the GraphComputer Memory mutates
with each iteration of a VertexProgram. We get more expressivity and faster
execution times. By "faster execution times," I'm suspecting that the 6 minute
Friendster job will drop down to 3 minutes. One less full scan of the dataset.
Thoughts?,
Marko.
http://markorodriguez.com
