mahout-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ted Dunning <>
Subject Re: Mahout on Google App Engine
Date Wed, 29 Sep 2010 19:11:39 GMT
I am not 100% up on the limitations imposed by the app engine, but I am sure
that you would need to swap out the persistence
layers for the recommendation engine.  That should not be difficult but is a
moderately low level task.

You will also be limited to deploying preconstructed classifier models or
only training for very limited amounts of time.  That would
be feasible using small batches of training data sent to the app engine at a
time.  Updating models from persistent storage in order
to allow them to be shared is doable, but would require new plumbing in
certain areas.

So I am not saying you have to move your customer facing app to EC2.  There
are limits that you need to investigate to determine
whether they impinge on your use and if you train any models off line, then
you need to do that training on something other than
App engine.

On Wed, Sep 29, 2010 at 11:21 AM, Sumi <> wrote:

> We are trying to prototype mahout cf for our application.
> Our app is gwt currently running on google app engine.
> Ted, are you saying that I need to move my application to EC2?
> On Tue, Sep 28, 2010 at 10:52 PM, Ted Dunning <>
> wrote:
> > I just noticed a posting on Nabble that I didn't ever see on our mailing
> > list.  Here is an answer.
> >
> > Damiano wrote:
> >
> >
> > Hi all,
> > I want to use Mahout in a distributed environment but I don't have CPUs
> to
> > put together and make a Hadoop cluster, so I would like to use Google App
> > Engine.
> > Is it possible to run Mahout on Google App Engine and take advantage of
> the
> > distributed environment?
> >
> > Thanks in advance,
> > Damiano Spina.
> >
> >
> > I don't think that this will work very well at all for much of Mahout.
>  You
> > can probably embed small classifier models in an AppEngine program, but
> the
> > AppEngine restricts file storage very severely so it is very unlikely
> that
> > you could run any training algorithms or recommendation engines.  For
> sure,
> > the experience would not be one suitable for a beginner.
> >
> > What I would recommend is that you put down a dollar or two and buy some
> > time from Amazon on EC2.  You can do some very large problems that way
> for
> > a
> > cost that is remarkably close to zero.
> >

  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message