[jira] [Commented] (GIRAPH-78) Be smarter about multiple instances of the same vertex

2011-11-16 Thread Jake Mannix (Commented) (JIRA)

[ 
https://issues.apache.org/jira/browse/GIRAPH-78?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13151873#comment-13151873
 ] 

Jake Mannix commented on GIRAPH-78:
---

Yeah, that's what I've been thinking too: each vertex has independent edge 
values to its destination, and doesn't keep a reference to the target vertex 
*value*, just its id.  So yeah, unless the  typed objects are big, I'm not 
sure what you can do here.

> Be smarter about multiple instances of the same vertex
> --
>
> Key: GIRAPH-78
> URL: https://issues.apache.org/jira/browse/GIRAPH-78
> Project: Giraph
>  Issue Type: Improvement
>Reporter: Jakob Homan
>
> In a graph such as 
> {noformat}a -> b, z
> b -> c, z
> c -> a, z
> ...
> z{noformat}
> where vertices a,b,c and are hosted on one worker and z is hosted on another, 
> it would be good to cache instances of z so a,b,c all point at the same 
> instance, rather than generating multiple copies of the same remote vertex 
> during vertex reading.  This is less important with primitive types and the 
> recent work done there, but very useful for more complex types.  Since the 
> vertex readers are in userland, it would be good to provide these facilities 
> as a library implementing users can access. ]

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: 
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira




[jira] [Commented] (GIRAPH-78) Be smarter about multiple instances of the same vertex

2011-11-16 Thread Avery Ching (Commented) (JIRA)

[ 
https://issues.apache.org/jira/browse/GIRAPH-78?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13151782#comment-13151782
 ] 

Avery Ching commented on GIRAPH-78:
---

Actually the more I think about it, this might not be too useful unless you 
have large vertexId objects.  I guess the idea would be to keep a cache, maybe 
in the GraphState or the WorkerContext.

> Be smarter about multiple instances of the same vertex
> --
>
> Key: GIRAPH-78
> URL: https://issues.apache.org/jira/browse/GIRAPH-78
> Project: Giraph
>  Issue Type: Improvement
>Reporter: Jakob Homan
>
> In a graph such as 
> {noformat}a -> b, z
> b -> c, z
> c -> a, z
> ...
> z{noformat}
> where vertices a,b,c and are hosted on one worker and z is hosted on another, 
> it would be good to cache instances of z so a,b,c all point at the same 
> instance, rather than generating multiple copies of the same remote vertex 
> during vertex reading.  This is less important with primitive types and the 
> recent work done there, but very useful for more complex types.  Since the 
> vertex readers are in userland, it would be good to provide these facilities 
> as a library implementing users can access. ]

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: 
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira




[jira] [Commented] (GIRAPH-78) Be smarter about multiple instances of the same vertex

2011-11-16 Thread Claudio Martella (Commented) (JIRA)

[ 
https://issues.apache.org/jira/browse/GIRAPH-78?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13151753#comment-13151753
 ] 

Claudio Martella commented on GIRAPH-78:


Yes, very nice, but how would you implement this? A caching Factory or you 
really want 100% re-use? That would require a per-worker index of Is.

> Be smarter about multiple instances of the same vertex
> --
>
> Key: GIRAPH-78
> URL: https://issues.apache.org/jira/browse/GIRAPH-78
> Project: Giraph
>  Issue Type: Improvement
>Reporter: Jakob Homan
>
> In a graph such as 
> {noformat}a -> b, z
> b -> c, z
> c -> a, z
> ...
> z{noformat}
> where vertices a,b,c and are hosted on one worker and z is hosted on another, 
> it would be good to cache instances of z so a,b,c all point at the same 
> instance, rather than generating multiple copies of the same remote vertex 
> during vertex reading.  This is less important with primitive types and the 
> recent work done there, but very useful for more complex types.  Since the 
> vertex readers are in userland, it would be good to provide these facilities 
> as a library implementing users can access. ]

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: 
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira




[jira] [Commented] (GIRAPH-78) Be smarter about multiple instances of the same vertex

2011-11-15 Thread Avery Ching (Commented) (JIRA)

[ 
https://issues.apache.org/jira/browse/GIRAPH-78?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13150673#comment-13150673
 ] 

Avery Ching commented on GIRAPH-78:
---

+1, this would be a great memory optimization.

> Be smarter about multiple instances of the same vertex
> --
>
> Key: GIRAPH-78
> URL: https://issues.apache.org/jira/browse/GIRAPH-78
> Project: Giraph
>  Issue Type: Improvement
>Reporter: Jakob Homan
>
> In a graph such as 
> {noformat}a -> b, z
> b -> c, z
> c -> a, z
> ...
> z{noformat}
> where vertices a,b,c and are hosted on one worker and z is hosted on another, 
> it would be good to cache instances of z so a,b,c all point at the same 
> instance, rather than generating multiple copies of the same remote vertex 
> during vertex reading.  This is less important with primitive types and the 
> recent work done there, but very useful for more complex types.  Since the 
> vertex readers are in userland, it would be good to provide these facilities 
> as a library implementing users can access. ]

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: 
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira