[
https://issues.apache.org/jira/browse/TINKERPOP3-613?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Marko A. Rodriguez updated TINKERPOP3-613:
------------------------------------------
Description:
It is important that reduce the size of the graph in OLAP so we can hold as
much in memory as possible and thus, decrease compute times. In order to reduce
the serialization cost, we have the following ticket TINKERPOP3-609.
This ticket is about reducing the in-memory footprint. Right now we use
TinkerGraph as the "star graph" for TinkerGraphComputer, GiraphGraphComputer,
and SparkGraphComputer. While TinkerGraph is small, it could be smaller.
We should introduce {{StarGraph}} that provides the following benefits over
{{TinkerGraph}}
* A {{StarEdge}} only has the adjacent vertex id.
* All labels are enum'd by some short number-based HashMap.
* {{StarElements}} should not maintain a reference back to Graph (i.e.
{{.graph()}}).
* The graph should only have a set of vertices, not edges.
* The above vertex set should be a bloom filter.
* There will be no indices in {{StarGraph}}.
* Removing one edge will automatically drop all edges (for fast MapReduce prep).
was:
It is important that reduce the size of the graph in OLAP so we can hold as
much in memory as possible and thus, decrease compute times. In order to reduce
the serialization cost, we have the following ticket TINKERPOP3-609.
This ticket is about reducing the in-memory footprint. Right now we use
TinkerGraph as the "star graph" for TinkerGraphComputer, GiraphGraphComputer,
and SparkGraphComputer. While TinkerGraph is small, it could be smaller.
We should introduce {{StarGraph}} that provides the following benefits over
{{TinkerGraph}}
* A {{StarEdge}} only has the adjacent vertex id.
* All labels are enum'd by some short number-based HashMap.
* {{StarElements}} should not maintain a reference back to Graph (i.e.
{{.graph()}}).
* The graph should only have a set of vertices, not edges.
* The above vertex set should be a bloom filter.
* There will be no indices in {{StarGraph}}.
> Create StarGraph as a micro "TinkerGraph"-style POJO graph.
> -----------------------------------------------------------
>
> Key: TINKERPOP3-613
> URL: https://issues.apache.org/jira/browse/TINKERPOP3-613
> Project: TinkerPop 3
> Issue Type: New Feature
> Components: process, structure
> Affects Versions: 3.0.0.GA
> Reporter: Marko A. Rodriguez
>
> It is important that reduce the size of the graph in OLAP so we can hold as
> much in memory as possible and thus, decrease compute times. In order to
> reduce the serialization cost, we have the following ticket TINKERPOP3-609.
> This ticket is about reducing the in-memory footprint. Right now we use
> TinkerGraph as the "star graph" for TinkerGraphComputer, GiraphGraphComputer,
> and SparkGraphComputer. While TinkerGraph is small, it could be smaller.
> We should introduce {{StarGraph}} that provides the following benefits over
> {{TinkerGraph}}
> * A {{StarEdge}} only has the adjacent vertex id.
> * All labels are enum'd by some short number-based HashMap.
> * {{StarElements}} should not maintain a reference back to Graph (i.e.
> {{.graph()}}).
> * The graph should only have a set of vertices, not edges.
> * The above vertex set should be a bloom filter.
> * There will be no indices in {{StarGraph}}.
> * Removing one edge will automatically drop all edges (for fast MapReduce
> prep).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
