giraph-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Eli Reisman (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (GIRAPH-157) Vertex to perform graph coloring on simple, connected, undirected graphs and related test.
Date Fri, 22 Jun 2012 16:40:43 GMT

    [ https://issues.apache.org/jira/browse/GIRAPH-157?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13399423#comment-13399423
] 

Eli Reisman commented on GIRAPH-157:
------------------------------------

I have asked a few people, and you're the first to answer, thanks! That makes a lot of sense.
The way I packaged it was more for testing than real use (similar to ConnectedComponentsVertex)
and I think the argument I for it in that sense is that the code shows how, when thinking
like a vertex, even hairy problems like this are easy for dull folks like myself to reason
about in a small amount of code. As far as scaling up to where Giraph wants to be, you're
exactly right. Point taken.

My other (selfish) hope was that others such as yourself might identify if, in fact, I oversimplified
the solution to the point where it wasn't a solution at all.

As far as time wasted, also a great point. But, my goal was to learn the application API for
Giraph (done) and get a good internship out of it (also done.) So, if this is all it yields
me and the patch withers here on the vine, thats OK with me.

I'm more concerned about your time wasted. Sorry for that. I appreciate your comments and
spending the time to have a dialogue with me on the topic!

If I might waste a bit more of it, it seems like graph coloring can have many applications
that don't need to scale the way other Giraph algorithms will to be somewhat useful (scheduling
classes/meetings for 10k students or a convention etc.) and while Giraph might be too much
horsepower for such uses, my thought was with Hadoop clusters cheap and everywhere now (my
school has a 64 node cluster) perhaps it could take a large ugly NP problem and just lower
the runtime a bit by leveraging a cluster since the algorithm is not going to be fast any
way you stack it up. I wouldn't get this thing near a social graph. etc.

What do you think? Is this silly? Go ahead and be honest you won't hurt my feelings. If the
algorithm is sound, the time is already wasted! Ever run across any graphs with a known chromatic
number in that size range?

Thanks again for talking to me about this, I really appreciate your input. I'll be working
on something more tangible in the coming weeks (lowering Giraph's memory profile per worker,
I hope) so perhaps I can redeem myself that way.


                
> Vertex to perform graph coloring on simple, connected, undirected graphs and related
test.
> ------------------------------------------------------------------------------------------
>
>                 Key: GIRAPH-157
>                 URL: https://issues.apache.org/jira/browse/GIRAPH-157
>             Project: Giraph
>          Issue Type: Test
>          Components: examples, test
>    Affects Versions: 0.2.0
>            Reporter: Eli Reisman
>            Assignee: Eli Reisman
>            Priority: Trivial
>              Labels: newbie
>         Attachments: GIRAPH-157-2.patch, GIRAPH-157.patch
>
>
> Hi. I am attempting to learn the Hadoop and Giraph codebases and wanted to write a simple
client application for Giraph to help me learn the ins and outs of it. This is a simple unit
test and vertex modeled after the ConnectedComponentsVertex and related test. The vertex test
runs whenever you run the "mvn test" or "mvn verify" suite of tests. When finished processing,
each vertex will have an integer value that is its color.
> This is a pretty simple implementation, and although I have tested it on a number of
small graphs of varied trickiness and it seems to rapidly arrive at a minimal coloring, its
hard (for me at least) to guess which possible coloring it will arrive at and I have no idea
how it will do on really big graphs yet without finding some more pre-colored larger test
graphs to try it on. Ideas anyone?
> Anyway, it was fun to put this together, and I'd be happy to improve it or receive some
help or advice to further the cause. Thanks again, I am hoping this will be the first of many
(hopefully more useful) contributions!
> Eli

--
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

        

Mime
View raw message