[
https://issues.apache.org/jira/browse/TINKERPOP-889?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
stephen mallette updated TINKERPOP-889:
---------------------------------------
Fix Version/s: (was: 3.2.7)
> Support for partitioned vertices in GraphComputer
> -------------------------------------------------
>
> Key: TINKERPOP-889
> URL: https://issues.apache.org/jira/browse/TINKERPOP-889
> Project: TinkerPop
> Issue Type: Improvement
> Components: process
> Affects Versions: 3.0.2-incubating
> Reporter: Matthias Broecheler
> Assignee: Marko A. Rodriguez
> Labels: breaking
>
> Most natural graphs have scale free distributions which means that some
> vertices in the graph have significantly more incident edges than others. On
> large graphs, it is therefore not uncommon to encounter vertices whose
> adjacency list is to large to be accommodated efficiently by a single machine
> (due to lack of sufficient memory or because they create a hotspot).
> Other graph computing frameworks have successfully addressed this "supernode
> problem" by partitioning the vertex's adjacency list and processing each
> subset of the adjacency list separately because merging the results (e.g.
> counting edges in each subset and then adding those values to get the total
> degree). TinkerPop should implement such functionality and allow partitioned
> vertex adjacency lists in the input to GraphComputer. This is a critical
> feature to make TP applicable to large graph computations.
> This can be implemented fairly easily for local messages using the message
> combiners. Global messages can be tricky however. See also TINKERPOP3-383.
> This has been partially implemented in Titan's Fulgora package
> (https://github.com/thinkaurelius/titan/tree/titan10/titan-core/src/main/java/com/thinkaurelius/titan/graphdb/olap).
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
