Toby DiPasquale on 5 Sep 2007 03:34:46 -0000


[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]

Re: [PhillyOnRails] jruby + hadoop?


On Tue, Sep 04, 2007 at 10:05:55PM -0400, Mat Schaffer wrote:
> So granted, I'm still reading the map/reduce paper, but this sorta  
> drove it home for me I think.  Thought I'd share it with those of us  
> not subscribed to Labnotes already.
> 
> http://blog.leetsoft.com/2007/9/2/futuretalk-couchdb
> (via http://blog.labnotes.org/2007/09/02/couchdb-thinking-beyond-the- 
> rdbms/)
> 
> Granted, it's not hadoop, but it's website-type data without an RDMBS  
> which put it in perspective pretty nicely.  I can sorta imagine this  
> becoming the norm and RDBMS becoming more of a niche market.

This is the sort of thing Hadoop is designed for. The guy who created
CouchDB is a long-time Lotus Notes guy and CouchDB is patterned after the
Notes/Domino database style/format. I really am interested in this project
and it seems really cool (I posted a link to this to Philly Lambda the
other day, since its written in Erlang) but I'm having trouble getting
people to understand the model on a conceptual level. I think people
either a) don't give a shit, or b) are so entrenched in the RDBMS way of
thinking that this just seems too alien to discern.

My suggestion would be to read the GFS paper first. About 40 times. Then
read the MapReduce paper. 40 times for that, too. Then think about how you
would do everything you're doing now if Larry had never stolen Ted's
technology from IBM and released it into the wilds of your brain. Really
think about it. Then go read a book or learn about a functional
programming language (Erlang, Haskell, etc).

I'll have some trivial and serious examples of what can be done in my talk
on Hadoop, if that's approved by the management (it is, after all, written
in Java and has nothing to do with Ruby or Rails). Most of the things that
you talk about re: Hadoop apply to CouchDB, as well, at least how its laid
out currently in the documentation. They also apply to append-only
flat-file storage architectures. They also apply to purely functional data
structures ("Hey, Erlang, I didn't see you come in... glad you could make
it.").

-- 
Toby DiPasquale
_______________________________________________
To unsubscribe or change your settings, visit:
http://lists.phillyonrails.org/mailman/listinfo/talk